Search

Monitoring variables such as dissolved oxygen, pH, and pond temperature is a key aspect of high-quality fish farming. Machine learning (ML) techniques have been proposed to model the dynamics of such variables to improve the fish farmer's decision-making. Most of the research on ML in aquaculture has focused on scenarios where devices for real-time data acquisition, storage, and remote monitoring are available, making it easy to develop accurate ML techniques. However, fish farmers do not necessarily have access to such devices. Many of them prefer to use equipment to manually measure these variables limiting the amount of available data to process. In this work, we study the use of random forests, multivariate linear regression, and artificial neural networks in scenarios with limited amount of measurements to analyze data from water-quality variables that are commonly measured in fish farming. We propose a methodology to build models in two scenarios: i) estimation of unobserved variables based on the observed ones, and ii) forecasting when a low amount of data is available for training. We show that random forests can be used to forecast dissolved oxygen, pond temperature, pH, ammonia, and ammonium when the water pond variables are measured only twice per day. Moreover, we showed that these prediction models can be implemented on a mobile-based information system and run in an average smartphone that fish farmers can afford. [ABSTRACT FROM AUTHOR]

The bandgap of two-dimensional (2D) materials plays an important role in their applications to various devices. For instance, the gapless nature of graphene limits the use of this material to semiconductor device applications, whereas the indirect bandgap of molybdenum disulfide is suitable for electrical and photo-device applications. Therefore, predicting the bandgap rapidly and accurately for a given 2D material structure has great scientific significance in the manufacturing of semiconductor devices. Compared to the extremely high computation cost of conventional first-principles calculations, machine learning (ML) based on statistics may be a promising alternative to predicting bandgaps. Although ML algorithms have been used to predict the properties of materials, they have rarely been used to predict the properties of 2D materials. In this study, we apply four ML algorithms to predict the bandgaps of 2D materials based on the computational 2D materials database (C2DB). Gradient boosted decision trees and random forests are more effective in predicting bandgaps of 2D materials with an R2 >90% and root-mean-square error (RMSE) of ~0.24 eV and 0.27 eV, respectively. By contrast, support vector regression and multi-layer perceptron show that R2 is >70% with RMSE of ~0.41 eV and 0.43 eV, respectively. Finally, when the bandgap calculated without spin-orbit coupling (SOC) is used as a feature, the RMSEs of the four ML models decrease greatly to 0.09 eV, 0.10 eV, 0.17 eV, and 0.12 eV, respectively. The R2 of all the models is >94%. These results show that the properties of 2D materials can be rapidly obtained by ML prediction with high precision. [ABSTRACT FROM AUTHOR]

Soil pH effects a wide range of critical biogeochemical processes that dictate plant growth and diversity. Previous literature has established the capacity of classification and regression trees (CARTs) to predict soil pH, but limitations of CARTs in this context have not been fully explored. The current study collected soil pH, climatic, and topographic data from 100 locations across New York's Temperate Deciduous Forests (in the United States of America) to investigate the extrapolative capacity of a previously developed CART model as compared to novel CART and random forest (RF) models. Results showed that the previously developed CART underperformed in terms of predictive accuracy (RRMSE = 14.52%) when compared to a novel tree (RRMSE = 9.33%), and that a novel random forest outperformed both models (RRMSE = 8.88%), though its predictions did not differ significantly from the novel tree (p = 0.26). The most important predictors for model construction were climatic factors. These findings confirm existing reports that CART models are constrained by the spatial autocorrelation of geographic data and encourage the restricted application of relevant machine learning models to regions from which training data was collected. They also contradict previous literature implying that random forests should meaningfully boost the predictive accuracy of CARTs in the context of soil pH. [ABSTRACT FROM AUTHOR]

Road crash fatality is a universal problem of the transportation system. A massive death toll caused annually due to road crash incidents, and among them, vulnerable road users (VRU) are endangered with high crash severity. This paper focuses on employing machine learning-based classification approaches for modelling injury severity of vulnerable road users—pedestrian, bicyclist, and motorcyclist. Specifically, this study aims to analyse critical features associated with different VRU groups—for pedestrian, bicyclist, motorcyclist and all VRU groups together. The critical factor of crash severity outcomes for these VRU groups is estimated in identifying the similarities and differences across different important features associated with different VRU groups. The crash data for the study is sourced from the state of Queensland in Australia for the years 2013 through 2019. The supervised machine learning algorithms considered for the empirical analysis includes the K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Forest (RF). In these models, 17 distinct road crash parameters are considered as input features to train models, which originate from road user characteristics, weather and environment, vehicle and driver condition, period, road characteristics and regions, traffic, and speed jurisdiction. These classification models are separately trained and tested for individual and unified VRU to assess crash severity levels. Afterwards, model performances are compared with each other to justify the best classifier where Random Forest classification models for all VRU modes are found to be comparatively robust in test accuracy: (motorcyclist: 72.30%, bicyclist: 64.45%, pedestrian: 67.23%, unified VRU: 68.57%). Based on the Random Forest model, the road crash features are ranked and compared according to their impact on crash severity classification. Furthermore, a model-based partial dependency of each road crash parameters on the severity levels is plotted and compared for each individual and unified VRU. This clarifies the tendency of road crash parameters to vary with different VRU crash severity. Based on the outcome of the comparative analysis, motorcyclists are found to be more likely exposed to higher crash severity, followed by pedestrians and bicyclists. [ABSTRACT FROM AUTHOR]

Background: Alcohol use disorder (AUD) is a chronic disease with a higher recurrence rate than that of other mental illnesses. Moreover, it requires continuous outpatient treatment for the patient to maintain abstinence. However, with a low probability of these patients to continue outpatient treatment, predicting and managing patients who might discontinue treatment becomes necessary. Accordingly, we developed a machine learning (ML) algorithm to predict which the risk of patients dropping out of outpatient treatment schemes. Methods: A total of 839 patients were selected out of 2,206 patients admitted for AUD in three hospitals under the Catholic Central Medical Center in Korea. We implemented six ML models—logistic regression, support vector machine, k-nearest neighbor, random forest, neural network, and AdaBoost—and compared the prediction performances thereof. Results: Among the six models, AdaBoost was selected as the final model for recommended use owing to its area under the receiver operating characteristic curve (AUROC) of 0.72. The four variables affecting the prediction based on feature importance were the length of hospitalization, age, residential area, and diabetes. Conclusion: An ML algorithm was developed herein to predict the risk of patients with AUD in Korea discontinuing outpatient treatment. By testing and validating various machine learning models, we determined the best performing model, AdaBoost, as the final model for recommended use. Using this model, clinicians can manage patients with high risks of discontinuing treatment and establish patient-specific treatment strategies. Therefore, our model can potentially enable patients with AUD to successfully complete their treatments by identifying them before they can drop out. [ABSTRACT FROM AUTHOR]

In the context of the current global pandemic and the limitations of the RT-PCR test, we propose a novel deep learning architecture, DFCN (Denoising Fully Connected Network). Since medical facilities around the world differ enormously in what laboratory tests or chest imaging may be available, DFCN is designed to be robust to missing input data. An ablation study extensively evaluates the performance benefits of the DFCN as well as its robustness to missing inputs. Data from 1088 patients with confirmed RT-PCR results are obtained from two independent medical facilities. The data includes results from 27 laboratory tests and a chest x-ray scored by a deep learning model. Training and test datasets are taken from different medical facilities. Data is made publicly available. The performance of DFCN in predicting the RT-PCR result is compared with 3 related architectures as well as a Random Forest baseline. All models are trained with varying levels of masked input data to encourage robustness to missing inputs. Missing data is simulated at test time by masking inputs randomly. DFCN outperforms all other models with statistical significance using random subsets of input data with 2-27 available inputs. When all 28 inputs are available DFCN obtains an AUC of 0.924, higher than any other model. Furthermore, with clinically meaningful subsets of parameters consisting of just 6 and 7 inputs respectively, DFCN achieves higher AUCs than any other model, with values of 0.909 and 0.919. [ABSTRACT FROM AUTHOR]

Identifying the factors controlling the spatial variability of soil metal elements could be a challenge task due to the interaction of environmental attributes and human activities. This study aimed to investigate the critical explanatory variables controlling total Ca, Cd, Cr, Cu, Zn, Fe, Mn, Mg, Pb, and Zn variations in the arable topsoil using classical statistics, principal component analysis, and random forest techniques. The work was conducted in the core region of the Three Gorges Reservoir of China. The explanatory variables included soil, topography, climate, vegetation, land use type, and distance-related parameters. Average concentrations of the metal elements were in order of Fe > Mg > Ca > Mn > Zn > Cr > Ni > Pb > Cu > Cd. Soil Cr, Fe, and Pb showed low variability while others presented medium variability. Average concentrations of Cr, Fe, Cd, and Mg exceeded their corresponding background values. There were highly positive correlations between all metal elements except Pb, Cd and Cr. The principal component analysis further demonstrated that the sources of Pb, Cd, and Cr differed with other elements. The results of random forest suggested that soil properties followed by topography were critical parameters affecting the variations of Ca, Mg, Mn, Fe, Ni, Zn, and Cu. Agricultural activities and soil properties were major factors controlling the variations of Pb, Cr, and Cd. Further study should be conducted to understand the relations between the metal elements and soil properties. [ABSTRACT FROM AUTHOR]

The aim of the current study was to investigate the relation between reticulorumen contractions and monitored cow behaviors. A purpose-built pressure measuring device was used and shown to be capable of detecting the known contraction patterns in the reticulorumen of four rumen-fistulated cows. Reticular pressure data was used to build a random forest algorithm, a learning algorithm based on a combination of decision trees, to detect rumination and other cow behaviors. In addition, we developed a peak-detection algorithm for rumination based on visual inspection of patterns in reticular pressure. Cow behaviors, differentiated in ruminating, eating, drinking, sleeping and 'other', as scored from video observation, were used to develop and test the algorithms. The results demonstrated that rumination of a cow can be detected by measuring pressure differences in the reticulum using either the random forest algorithm or the peak-detection algorithm. The random forest algorithm showed very robust performances for detecting rumination with an accuracy of 0.98, a sensitivity of 0.95 and a specificity of 0.99. The peak-detection algorithm could detect rumination robustly, with an accuracy of 0.92, a sensitivity of 0.97 and a specificity of 0.90. In addition, we provide proof of principle that a random forest algorithm can also detect eating, drinking and sleeping behavior from the same data with performances above 0.90 for all measures. The measurement device used in this study needed rumen-fistulated cows, but the results indicate that behavior detection using algorithms based on only measurements in the reticulum is feasible. This is promising as it may allow future wireless sensor techniques in the reticulum to continuously monitor a range of important behaviors of cows. [ABSTRACT FROM AUTHOR]

Aim: The aim was to predict and understand variations in swimmer performance between individual and relay events, and develop a predictive model for the 4x200-m swimming freestyle relay event to help inform team selection and strategy. Data and methods: Race data for 716 relay finals (4 x 200-m freestyle) from 14 international competitions between 2010–2018 were analysed. Individual 200-m freestyle season best time for the same year was located for each swimmer. Linear regression and machine learning was applied to 4 x 200-m swimming freestyle relay events. Results: Compared to the individual event, the lowest ranked swimmer in the team (-0.62 s, CI = [−0.94, −0.30]) and American swimmers (−0.48 s [−0.89, −0.08]) typically swam faster 200-m times in relay events. Random forest models predicted gold, silver, bronze and non-medal with 100%, up to 41%, up to 63%, and 93% sensitivity, respectively. Discussion: Team finishing position was strongly associated with the differential time to the fastest team (mean decrease in Gini (MDG) when this variable was omitted = 31.3), world rankings of team members (average ranking MDG of 18.9), and the order of swimmers (MDG = 6.9). Differential times are based on the sum of individual swimmer's season's best times, and along with world rankings, reflect team strength. In contrast, the order of swimmers reflects strategy. This type of analysis could assist coaches and support staff in selecting swimmers and team orders for relay events to enhance the likelihood of success. [ABSTRACT FROM AUTHOR]

Due to difficulty in early diagnosis of Alzheimer's disease (AD) related to cost and differentiated capability, it is necessary to identify low-cost, accessible, and reliable tools for identifying AD risk in the preclinical stage. We hypothesized that cognitive ability, as expressed in the vocal features in daily conversation, is associated with AD progression. Thus, we have developed a novel machine learning prediction model to identify AD risk by using the rich voice data collected from daily conversations, and evaluated its predictive performance in comparison with a classification method based on the Japanese version of the Telephone Interview for Cognitive Status (TICS-J). We used 1,465 audio data files from 99 Healthy controls (HC) and 151 audio data files recorded from 24 AD patients derived from a dementia prevention program conducted by Hachioji City, Tokyo, between March and May 2020. After extracting vocal features from each audio file, we developed machine-learning models based on extreme gradient boosting (XGBoost), random forest (RF), and logistic regression (LR), using each audio file as one observation. We evaluated the predictive performance of the developed models by describing the receiver operating characteristic (ROC) curve, calculating the areas under the curve (AUCs), sensitivity, and specificity. Further, we conducted classifications by considering each participant as one observation, computing the average of their audio files' predictive value, and making comparisons with the predictive performance of the TICS-J based questionnaire. Of 1,616 audio files in total, 1,308 (81.0%) were randomly allocated to the training data and 308 (19.1%) to the validation data. For audio file-based prediction, the AUCs for XGboost, RF, and LR were 0.863 (95% confidence interval [CI]: 0.794–0.931), 0.882 (95% CI: 0.840–0.924), and 0.893 (95%CI: 0.832–0.954), respectively. For participant-based prediction, the AUC for XGboost, RF, LR, and TICS-J were 1.000 (95%CI: 1.000–1.000), 1.000 (95%CI: 1.000–1.000), 0.972 (95%CI: 0.918–1.000) and 0.917 (95%CI: 0.918–1.000), respectively. There was difference in predictive accuracy of XGBoost and TICS-J with almost approached significance (p = 0.065). Our novel prediction model using the vocal features of daily conversations demonstrated the potential to be useful for the AD risk assessment. [ABSTRACT FROM AUTHOR]

While it is generally accepted that language and speech have genetic foundations, and that the widespread inter-individual variation observed in many of their aspects is partly driven by variation in genes, it is much less clear if differences between languages may also be partly rooted in our genes. One such proposal is that the population frequencies of the so-called "derived" alleles of two genes involved in brain growth and development, ASPM and Microcephalin, are related to the probability of speaking a tone language or not. The original study introducing this proposal used a cross-linguistic statistical approach, showing that these associations are "special" when compared with many other possible relationships between genetic variants and linguistic features. Recent experimental evidence supports strongly a negative effect of the "derived" allele of ASPM on tone perception and/or processing within individuals, but failed to find any effect for Microcephalin. Motivated by these experimental findings, I conduct here a cross-linguistic statistical test, using a larger and updated dataset of 175 samples from 129 unique (meta)populations, and a battery of methods including mixed-effects regression (Bayesian and maximum-likelihood), mediation and path analysis, decision trees and random forests, using permutations and restricted sampling to control for the confounding effects of genealogy (language families) and contact (macroareas). Overall, the results support a negative weak effect of ASPM-D against the presence of tone above and beyond the strong confounding influences of genealogy and contact, but they suggest that the original association between tone and MCPH1 might have been a false positive, explained by differences between populations and languages within and outside Africa. Thus, these cross-linguistic population-scale statistical results are fully consonant with the inter-individual-level experimental results, and suggest that the observed linguistic diversity may be, at least in some cases, partly driven by genetic diversity. [ABSTRACT FROM AUTHOR]

Background: Deep neural networks learn from former experiences on a large scale and can be used to predict future disease activity as potential clinical decision support. AdaptiveNet is a novel adaptive recurrent neural network optimized to deal with heterogeneous and missing clinical data. Objective: We investigate AdaptiveNet for the prediction of individual disease activity in patients from a rheumatoid arthritis (RA) registry. Methods: Demographic and disease characteristics from over 9500 patients and 65.000 visits from the Swiss Quality Management (SCQM) database were used to train and evaluate the network. Patient characteristics, clinical and patient reported outcomes, laboratory values and medication were used as input features. DAS28-BSR served as a target to predict active RA and future numeric individual disease activity by classification and regression. Results: AdaptiveNet predicted active disease defined as DAS28-BSR >2.6 at the next visit with an overall accuracy of 75.6% (SD +- 0.7%) and a sensitivity and specificity of 84.2% (SD +- 1.6%) and 61.5% (SD +- 3.6%), respectively. Prediction performance was significantly higher in patients with a disease duration >3 years and positive rheumatoid factor. Regression allowed forecasting individual DAS28-BSR values with a mean squared error (MSE) of 0.9 (SD +- 0.05). This corresponds to a 8% deviation between estimated and real DAS28-BSR values. Compared to linear regression, random forest and support vector machines, AdaptiveNet showed an increased performance of over 7% in MSE. Medication played a minor role in the prediction of RA disease activity. Conclusion: AdaptiveNet has a superior capacity to predict numeric RA disease activity compared to classical machine learning architectures. All investigated models had limitations in low specificity. [ABSTRACT FROM AUTHOR]

Aims: Malnutrition is a major health issue among Bangladeshi under-five (U5) children. Children are malnourished if the calories and proteins they take through their diet are not sufficient for their growth and maintenance. The goal of the research was to use machine learning (ML) algorithms to detect the risk factors of malnutrition (stunted, wasted, and underweight) as well as their prediction. Methods: This work utilized malnutrition data that was derived from Bangladesh Demographic and Health Survey which was conducted in 2014. The selected dataset consisted of 7079 children with 13 factors. The potential risks of malnutrition have been identified by logistic regression (LR). Moreover, 3 ML classifiers (support vector machine (SVM), random forest (RF), and LR) have been implemented for predicting malnutrition and the performance of these ML algorithms were assessed on the basis of accuracy. Results: The average prevalence of stunted, wasted, and underweight was 35.4%, 15.4%, and 32.8%, respectively. It was noted that LR identified five risk factors for stunting and underweight, as well as four factors for wasting. Results illustrated that RF can be accurately classified as stunted, wasted, and underweight children and obtained the highest accuracy of 88.3% for stunted, 87.7% for wasted, and 85.7% for underweight. Conclusion: This research focused on the identification and prediction of major risk factors for stunting, wasting, and underweight using ML algorithms which will aid policymakers in reducing malnutrition among Bangladesh's U5 children. [ABSTRACT FROM AUTHOR]

Sarcasm is the main reason behind the faulty classification of tweets. It brings a challenge in natural language processing (NLP) as it hampers the method of finding people's actual sentiment. Various feature engineering techniques are being investigated for the automatic detection of sarcasm. However, most related techniques have always concentrated only on the content-based features in sarcastic expression, leaving the contextual information in isolation. This leads to a loss of the semantics of words in the sarcastic expression. Another drawback is the sparsity of the training data. Due to the word limit of microblog, the feature vector's values for each sample constructed by BoW produces null features. To address the above-named problems, a Multi-feature Fusion Framework is proposed using two classification stages. The first stage classification is constructed with the lexical feature only, extracted using the BoW technique, and trained using five standard classifiers, including SVM, DT, KNN, LR, and RF, to predict the sarcastic tendency. In stage two, the constructed lexical sarcastic tendency feature is fused with eight other proposed features for modelling a context to obtain a final prediction. The effectiveness of the developed framework is tested with various experimental analysis to obtain classifiers' performance. The evaluation shows that our constructed classification models based on the developed novel feature fusion obtained results with a precision of 0.947 using a Random Forest classifier. Finally, the obtained results were compared with the results of three baseline approaches. The comparison outcome shows the significance of the proposed framework. [ABSTRACT FROM AUTHOR]

Quantification of endogenous biomarkers in clinical studies requires careful evaluation of a number of assay performance parameters. Comparisons of absolute values from several clinical studies can enable retrospective analyses further elucidating the biology of a given biomarker across various study populations. We characterized the performance of a highly multiplex bioanalytical method for quantification of phosphatidylinositols (PI). Hydrophilic interaction chromatography (HILIC) and multiple reaction monitoring (MRM) were employed for targeted multiplex quantification. Odd-chain PI species that are not normally present in human plasma were utilized as surrogate analytes (SA) to assess various assay performance parameters and establish a definitive dynamic linear range for PI lipids. To correct for batch effects, Systematic Error Removal using Random Forest (SERRF) normalization algorithm was employed and used to bridge raw values between two clinical studies, enabling quantitative comparison of their absolute values. A high throughput method was developed, qualified, transferred to an automation platform and applied to sample testing in two clinical trials in healthy volunteers (NCT03001297) and stable Coronary Artery Disease (CAD, NCT03351738) subjects. The method demonstrated acceptable precision and accuracy (±30%) over linear range of 1–1000 nM for SA and 8-fold dilutional linearity for endogenous PI. We determined that mean-adjusted average QC performed best for normalization using SERRF. The comparison of two studies revealed that healthy subject levels of PI are consistently higher across PI species compared to CAD subjects identifying a potential lipid biomarker to be explored in future studies. [ABSTRACT FROM AUTHOR]

China became the country with the largest global carbon emissions in 2007. Cities are regional population and economic centers and are the main sources of carbon emissions. However, factors influencing carbon emissions from cities can vary with geographic location and the development history of the cities, rendering it difficult to explicitly quantify the influence of individual factors on carbon emissions. In this study, random forest (RF) machine learning algorithms were applied to analyze the relationships between factors and carbon emissions in cities using real-world data from Chinese cities. Seventy-three cities in three urban agglomerations within the Yangtze River Economic Belt were evaluated with respect to urban carbon emissions using data from regional energy balance tables for the years 2000, 2007, 2012, and 2017. The RF algorithm was then used to select 16 prototypical cities based on 10 influencing factors that affect urban carbon emissions while considering five primary factors: population, industry, technology levels, consumption, and openness to the outside world. Subsequently, 18 consecutive years of data from 2000 to 2017 were used to construct RFs to investigate the temporal predictability of carbon emission variation in the 16 cities based on regional differences. Results indicated that the RF approach is a practical tool to study the connection between various influencing factors and carbon emissions in the Yangtze River Economic Belt from different perspectives. Furthermore, regional differences among the primary carbon emission influencing factors for each city were clearly observed and were related to urban population characteristics, urbanization level, industrial structures, and degree of openness to the outside world. These factors variably affected different cities, but the results indicate that regional emission reductions have achieved positive results, with overall simulation trends shifting from underestimation to overestimation of emissions. [ABSTRACT FROM AUTHOR]

Objective: This study aimed to develop and validate a claims-based, machine learning algorithm to predict clinical outcomes across both medical and surgical patient populations. Methods: This retrospective, observational cohort study, used a random 5% sample of 770,777 fee-for-service Medicare beneficiaries with an inpatient hospitalization between 2009–2011. The machine learning algorithms tested included: support vector machine, random forest, multilayer perceptron, extreme gradient boosted tree, and logistic regression. The extreme gradient boosted tree algorithm outperformed the alternatives and was the machine learning method used for the final risk model. Primary outcome was 30-day mortality. Secondary outcomes were: rehospitalization, and any of 23 adverse clinical events occurring within 30 days of the index admission date. Results: The machine learning algorithm performance was evaluated by both the area under the receiver operating curve (AUROC) and Brier Score. The risk model demonstrated high performance for prediction of: 30-day mortality (AUROC = 0.88; Brier Score = 0.06), and 17 of the 23 adverse events (AUROC range: 0.80–0.86; Brier Score range: 0.01–0.05). The risk model demonstrated moderate performance for prediction of: rehospitalization within 30 days (AUROC = 0.73; Brier Score: = 0.07) and six of the 23 adverse events (AUROC range: 0.74–0.79; Brier Score range: 0.01–0.02). The machine learning risk model performed comparably on a second, independent validation dataset, confirming that the risk model was not overfit. Conclusions and relevance: We have developed and validated a robust, claims-based, machine learning risk model that is applicable to both medical and surgical patient populations and demonstrates comparable predictive accuracy to existing risk models. [ABSTRACT FROM AUTHOR]

Biological pest control (i.e. 'biocontrol') agents can have direct and indirect non-target impacts, and predicting these effects (especially indirect impacts) remains a central challenge in biocontrol risk assessment. The analysis of ecological networks offers a promising approach to understanding the community-wide impacts of biocontrol agents (via direct and indirect interactions). Independently, species traits and phylogenies have been shown to successfully predict species interactions and network structure (alleviating the need to collect quantitative interaction data), but whether these approaches can be combined to predict indirect impacts of natural enemies remains untested. Whether predictions of interactions (i.e. direct effects) can be made equally well for generalists vs. specialists, abundant vs. less abundant species, and across different habitat types is also untested for consumer-prey interactions. Here, we used two machine-learning techniques (random forest and k-nearest neighbour; KNN) to test whether we could accurately predict empirically-observed quantitative host-parasitoid networks using trait and phylogenetic information. Then, we tested whether the accuracy of machine-learning-predicted interactions depended on the generality or abundance of the interacting partners, or on the source (habitat type) of the training data. Finally, we used these predicted networks to generate predictions of indirect effects via shared natural enemies (i.e. apparent competition), and tested these predictions against empirically observed indirect effects between hosts. We found that random-forest models predicted host-parasitoid pairwise interactions (which could be used to predict attack of non-target host species) more successfully than KNN. This predictive ability depended on the generality of the interacting partners for KNN models, and depended on species' abundances for both random-forest and KNN models, but did not depend on the source (habitat type) of data used to train the models. Further, although our machine-learning informed methods could significantly predict indirect effects, the explanatory power of our machine-learning models for indirect interactions was reasonably low. Combining machine-learning and network approaches provides a starting point for reducing risk in biocontrol introductions, and could be applied more generally to predicting species interactions such as impacts of invasive species. [ABSTRACT FROM AUTHOR]

The objective of the current study was to build predictive models for suicidal ideation in a sample of children aged 9–10 using features previously implicated in risk among older adolescent and adult populations. This case-control analysis utilized baseline data from the Adolescent Brain and Cognitive Development (ABCD) Study, collected from 21 research sites across the United States (N = 11,369). Several regression and ensemble learning models were compared on their ability to classify individuals with suicidal ideation and/or attempt from healthy controls, as assessed by the Kiddie Schedule for Affective Disorders and Schizophrenia–Present and Lifetime Version. When comparing control participants (mean age: 9.92±0.62 years; 4944 girls [49%]) to participants with suicidal ideation (mean age: 9.89±0.63 years; 451 girls [40%]), both logistic regression with feature selection and elastic net without feature selection predicted suicidal ideation with an AUC of 0.70 (CI 95%: 0.70–0.71). The random forest with feature selection trained to predict suicidal ideation predicted a holdout set of children with a history of suicidal ideation and attempt (mean age: 9.96±0.62 years; 79 girls [41%]) from controls with an AUC of 0.77 (CI 95%: 0.76–0.77). Important features from these models included feelings of loneliness and worthlessness, impulsivity, prodromal psychosis symptoms, and behavioral problems. This investigation provided an unprecedented opportunity to identify suicide risk in youth. The use of machine learning to examine a large number of predictors spanning a variety of domains provides novel insight into transdiagnostic factors important for risk classification. [ABSTRACT FROM AUTHOR]

Donor-Recipient (D-R) matching is one of the main challenges to be fulfilled nowadays. Due to the increasing number of recipients and the small amount of donors in liver transplantation, the allocation method is crucial. In this paper, to establish a fair comparison, the United Network for Organ Sharing database was used with 4 different end-points (3 months, and 1, 2 and 5 years), with a total of 39, 189 D-R pairs and 28 donor and recipient variables. Modelling techniques were divided into two groups: 1) classical statistical methods, including Logistic Regression (LR) and Naïve Bayes (NB), and 2) standard machine learning techniques, including Multilayer Perceptron (MLP), Random Forest (RF), Gradient Boosting (GB) or Support Vector Machines (SVM), among others. The methods were compared with standard scores, MELD, SOFT and BAR. For the 5-years end-point, LR (AUC = 0.654) outperformed several machine learning techniques, such as MLP (AUC = 0.599), GB (AUC = 0.600), SVM (AUC = 0.624) or RF (AUC = 0.644), among others. Moreover, LR also outperformed standard scores. The same pattern was reproduced for the others 3 end-points. Complex machine learning methods were not able to improve the performance of liver allocation, probably due to the implicit limitations associated to the collection process of the database. [ABSTRACT FROM AUTHOR]

Purpose: Abnormalities of the running pattern of choroidal vessel have been reported in eyes with pachychoroid diseases. However, it is difficult for clinicians to judge the running pattern with high reproducibility. Thus, the purpose of this study was to compare the degree of concordance of the running pattern of the choroidal vessels between that determined by artificial intelligence (AI) to that determined by experienced clinicians. Methods: The running pattern of the choroidal vessels in en face images of Haller's layer of 413 normal and pachychoroid diseased eyes was classified as symmetrical or asymmetrical by human raters and by three supervised machine learning models; the support vector machine (SVM), Xception, and random forest models. The data from the human raters were used as the supervised data. The accuracy rates of the human raters and the certainty of AI's answers were compared using confidence scores (CSs). Results: The choroidal vascular running pattern could be determined by each AI model with an area under the curve better than 0.94. The random forest method was able to discriminate with the highest accuracy among the three AIs. In the CS analyses, the percentage of certainty was highest (66.4%) and that of uncertainty was lowest (6.1%) in the agreement group. On the other hand, the rate of uncertainty was highest (27.3%) in the disagreement group. Conclusion: AI algorithm can automatically classify with ambiguous criteria the presence or absence of a symmetrical blood vessel running pattern of the choroid. The classification was as good as that of supervised humans in accuracy and reproducibility. [ABSTRACT FROM AUTHOR]

Wisconsin and Quebec are the world leading cranberry-producing regions. Cranberries are grown in acidic, naturally low-fertility sandy beds. Cranberry fertilization is guided by general soil and tissue nutrient tests in addition to yield target and vegetative biomass. However, other factors such as cultivar, location, and carbon and nutrient storage impact cranberry nutrition and yield. The objective of this study was to customize nutrient diagnosis and fertilizer recommendation at local scale and for next-year cranberry production after accounting for local factors and carbon and nutrient carryover effects. We collected 1768 observations from on-farm surveys and fertilizer trials in Quebec and Wisconsin to elaborate a machine learning model using minimum datasets. We tested carryover effects in a 5-year Quebec fertilizer experiment established on permanent plots. Micronutrients contributed more than macronutrients to variation in tissue compositions. Random Forest model related accurately current-year berry yield to location, cultivars, climatic indices, fertilization, and tissue and soil tests as features (classification accuracy of 0.83). Comparing compositions of defective and successful tissue compositions in the Euclidean space of tissue compositions, the general across-factor diagnosis differed from the local factor-specific diagnosis. Nutrient standards elaborated in one region could hardly be transposed to another and, within the same region, from one bed to another due to site-specific characteristics. Next-year yield and nutrient adjustment could be predicted accurately from current-year yield and tissue composition and other features, with R2 value of 0.73 in regression mode and classification accuracy of 0.85. Compositional and machine learning methods proved to be effective to customize nutrient diagnosis and predict site-specific measures for nutrient management of cranberry stands. This study emphasized the need to acquire large experimental and observational datasets to capture the numerous factor combinations impacting current and next-year cranberry yields at local scale. [ABSTRACT FROM AUTHOR]

Background: Occupational stress is associated with adverse outcomes for medical professionals and patients. In our cross-sectional study with 136 general practices, 26.4% of 550 practice assistants showed high chronic stress. As machine learning strategies offer the opportunity to improve understanding of chronic stress by exploiting complex interactions between variables, we used data from our previous study to derive the best analytic model for chronic stress: four common machine learning (ML) approaches are compared to a classical statistical procedure. Methods: We applied four machine learning classifiers (random forest, support vector machine, K-nearest neighbors', and artificial neural network) and logistic regression as standard approach to analyze factors contributing to chronic stress in practice assistants. Chronic stress had been measured by the standardized, self-administered TICS-SSCS questionnaire. The performance of these models was compared in terms of predictive accuracy based on the 'operating area under the curve' (AUC), sensitivity, and positive predictive value. Findings: Compared to the standard logistic regression model (AUC 0.636, 95% CI 0.490–0.674), all machine learning models improved prediction: random forest +20.8% (AUC 0.844, 95% CI 0.684–0.843), artificial neural network +12.4% (AUC 0.760, 95% CI 0.605–0.777), support vector machine +15.1% (AUC 0.787, 95% CI 0.634–0.802), and K-nearest neighbours +7.1% (AUC 0.707, 95% CI 0.556–0.735). As best prediction model, random forest showed a sensitivity of 99% and a positive predictive value of 79%. Using the variable frequencies at the decision nodes of the random forest model, the following five work characteristics influence chronic stress: too much work, high demand to concentrate, time pressure, complicated tasks, and insufficient support by practice leaders. Conclusions: Regarding chronic stress prediction, machine learning classifiers, especially random forest, provided more accurate prediction compared to classical logistic regression. Interventions to reduce chronic stress in practice personnel should primarily address the identified workplace characteristics. [ABSTRACT FROM AUTHOR]

Time-to-event analysis is a common occurrence in political science. In recent years, there has been an increased usage of machine learning methods in quantitative political science research. This article advocates for the implementation of machine learning duration models to assist in a sound model selection process. We provide a brief tutorial introduction to the random survival forest (RSF) algorithm and contrast it to a popular predecessor, the Cox proportional hazards model, with emphasis on methodological utility for political science researchers. We implement both methods for simulated time-to-event data and the Power-Sharing Event Dataset (PSED) to assist researchers in evaluating the merits of machine learning duration models. We provide evidence of significantly higher survival probabilities for peace agreements with 3rd party mediated design and implementation. We also detect increased survival probabilities for peace agreements that incorporate territorial power-sharing and avoid multiple rebel party signatories. Further, the RSF, a previously under-used method for analyzing political science time-to event data, provides a novel approach for ranking of peace agreement criteria importance in predicting peace agreement duration. Our findings demonstrate a scenario exhibiting the interpretability and performance of RSF for political science time-to-event data. These findings justify the robust interpretability and competitive performance of the random survival forest algorithm in numerous circumstances, in addition to promoting a diverse, holistic model-selection process for time-to-event political science data. [ABSTRACT FROM AUTHOR]

Alzheimer's Disease (AD) conversion prediction from the mild cognitive impairment (MCI) stage has been a difficult challenge. This study focuses on providing an individualized MCI to AD conversion prediction using a balanced random forest model that leverages clinical data. In order to do this, 383 Early Mild Cognitive Impairment (EMCI) patients were gathered from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Of these patients, 49 would eventually convert to AD (EMCI_C), whereas the remaining 334 did not convert (EMCI_NC). All of these patients were split randomly into training and testing data sets with 95 patients reserved for testing. Nine clinical features were selected, comprised of a mix of demographic, brain volume, and cognitive testing variables. Oversampling was then performed in order to balance the initially imbalanced classes prior to training the model with 1000 estimators. Our results showed that a random forest model was effective (93.6% accuracy) at predicting the conversion of EMCI patients to AD based on these clinical features. Additionally, we focus on explainability by assessing the importance of each clinical feature. Our model could impact the clinical environment as a tool to predict the conversion to AD from a prodromal stage or to identify ideal candidates for clinical trials. [ABSTRACT FROM AUTHOR]

Accurate prediction of cancer stage is important in that it enables more appropriate treatment for patients with cancer. Many measures or methods have been proposed for more accurate prediction of cancer stage, but recently, machine learning, especially deep learning-based methods have been receiving increasing attention, mostly owing to their good prediction accuracy in many applications. Machine learning methods can be applied to high throughput DNA mutation or RNA expression data to predict cancer stage. However, because the number of genes or markers generally exceeds 10,000, a considerable number of data samples is required to guarantee high prediction accuracy. To solve this problem of a small number of clinical samples, we used a Generative Adversarial Networks (GANs) to augment the samples. Because GANs are not effective with whole genes, we first selected significant genes using DNA mutation data and random forest feature ranking. Next, RNA expression data for selected genes were expanded using GANs. We compared the classification accuracies using original dataset and expanded datasets generated by proposed and existing methods, using random forest, Deep Neural Networks (DNNs), and 1-Dimensional Convolutional Neural Networks (1DCNN). When using the 1DCNN, the F1 score of GAN5 (a 5-fold increase in data) was improved by 39% in relation to the original data. Moreover, the results using only 30% of the data were better than those using all of the data. Our attempt is the first to use GAN for augmentation using numeric data for both DNA and RNA. The augmented datasets obtained using the proposed method demonstrated significantly increased classification accuracy for most cases. By using GAN and 1DCNN in the prediction of cancer stage, we confirmed that good results can be obtained even with small amounts of samples, and it is expected that a great deal of the cost and time required to obtain clinical samples will be reduced. The proposed sample augmentation method could also be applied for other purposes, such as prognostic prediction or cancer classification. [ABSTRACT FROM AUTHOR]

Fall-risk classification is a challenging but necessary task to enable the recommendation of preventative programs for individuals identified at risk for falling. Existing research has primarily focused on older adults, with no predictive fall-risk models for lower limb amputees, despite their greater likelihood of fall-risk than older adults. In this study, 89 amputees with varying degrees of lower limb amputation were asked if they had fallen in the past 6 months. Those who reported at least one fall were considered a fall risk. Each participant performed a 6 minute walk test (6MWT) with an Android smartphone placed in a holder located on the back of the pelvis. A fall-risk classification method was developed using data from sensors within the smartphone. The Ottawa Hospital Rehabilitation Center Walk Test app captured accelerometer and gyroscope data during the 6MWT. From this data, foot strikes were identified, and 248 features were extracted from the collection of steps. Steps were segmented into turn and straight walking, and four different data sets were created: turn steps, straightaway steps, straightaway and turn steps, and all steps. From these, three feature selection techniques (correlation-based feature selection, relief F, and extra trees classifier ensemble) were used to eliminate redundant or ineffective features. Each feature subset was tested with a random forest classifier and optimized for the best number of trees. The best model used turn data, with three features selected by Correlation-based feature selection (CFS), and used 500 trees in a random forest classifier. The resulting metrics were 81.3% accuracy, 57.2% sensitivity, 94.9% specificity, a Matthews correlation coefficient of 0.587, and an F1 score of 0.83. Since the outcomes are comparable to metrics achieved by existing clinical tests, the classifier may be viable for use in clinical practice. [ABSTRACT FROM AUTHOR]

Prior research showed that there is a growing consensus among researchers, which point out a key role of external knowledge sources such as external R&D and technologies in enhancing firms´ innovation. However, firms´ from catching-up Central and Eastern European (CEE) countries have already shown in the past that their innovation models differ from those applied, for example, in Western Europe. This study therefore introduces a novel two-staged model combining artificial neural networks and random forests to reveal the importance of internal and external factors influencing firms´ innovation performance in the case of 3,361 firms from six catching-up CEE countries (Czech Republic, Slovakia, Poland, Estonia, Latvia and Lithuania), by using the World Banks´ Enterprise Survey data from 2019. We confirm the hypothesis that innovators in the catching-up CEE countries depend more on internal knowledge sources and, moreover, that participation in the firms groups represents an important factor of firms´ innovation. Surprisingly, we reject the hypothesis that foreign technologies are a crucial source of external knowledge. This study contributes to the theories of open innovation and absorptive capacity in the context of selected CEE countries and provides several practical implications for firms. [ABSTRACT FROM AUTHOR]

Objective: To assess whether machine learning algorithms (MLA) can predict eyes that will undergo rapid glaucoma progression based on an initial visual field (VF) test. Design: Retrospective analysis of longitudinal data. Subjects: 175,786 VFs (22,925 initial VFs) from 14,217 patients who completed ≥5 reliable VFs at academic glaucoma centers were included. Methods: Summary measures and reliability metrics from the initial VF and age were used to train MLA designed to predict the likelihood of rapid progression. Additionally, the neural network model was trained with point-wise threshold data in addition to summary measures, reliability metrics and age. 80% of eyes were used for a training set and 20% were used as a test set. MLA test set performance was assessed using the area under the receiver operating curve (AUC). Performance of models trained on initial VF data alone was compared to performance of models trained on data from the first two VFs. Main outcome measures: Accuracy in predicting future rapid progression defined as MD worsening more than 1 dB/year. Results: 1,968 eyes (8.6%) underwent rapid progression. The support vector machine model (AUC 0.72 [95% CI 0.70–0.75]) most accurately predicted rapid progression when trained on initial VF data. Artificial neural network, random forest, logistic regression and naïve Bayes classifiers produced AUC of 0.72, 0.70, 0.69, 0.68 respectively. Models trained on data from the first two VFs performed no better than top models trained on the initial VF alone. Based on the odds ratio (OR) from logistic regression and variable importance plots from the random forest model, older age (OR: 1.41 per 10 year increment [95% CI: 1.34 to 1.08]) and higher pattern standard deviation (OR: 1.31 per 5-dB increment [95% CI: 1.18 to 1.46]) were the variables in the initial VF most strongly associated with rapid progression. Conclusions: MLA can be used to predict eyes at risk for rapid progression with modest accuracy based on an initial VF test. Incorporating additional clinical data to the current model may offer opportunities to predict patients most likely to rapidly progress with even greater accuracy. [ABSTRACT FROM AUTHOR]

We propose an agent-based model for predicting individual flight delays in an entire air traffic network. In contrast to previous work, more detailed parameter estimation methods were incorporated into the agent-based model, acting on the state transitions of agents. Specifically, a conditional probability model was proposed for modifying the expected departure time, which was used to indicate whether a flight had experienced the necessary waiting due to Ground Delay Programs (GDPs) or carrier-related reasons. Additionally, two random forest regression models were presented for estimating the turnaround time and the elapsed time of flight agents in the agent-based delay prediction model. The parameter models were trained and fitted using the flight data for 2017 in the United States. The performance of the delay prediction model was tested for thirty days with three types of delay levels (low, medium, and high), which were randomly selected from 2018. The experimental results showed that the average absolute error in the test days was 6.8 min, and the classification accuracy with a 15 min threshold for a two-hour forecast horizon was 89.5%. The performance of our model outperformed that of existing research. Additionally, the positive effect of introducing parameter models and the negative impact of increasing the prediction horizon on the prediction performance were further studied. [ABSTRACT FROM AUTHOR]

Web-based innovation indicators may provide new insights into firm-level innovation activities. However, little is known yet about the accuracy and relevance of web-based information for measuring innovation. In this study, we use data on 4,487 firms from the Mannheim Innovation Panel (MIP) 2019, the German contribution to the European Community Innovation Survey (CIS), to analyze which website characteristics perform as predictors of innovation activity at the firm level. Website characteristics are measured by several data mining methods and are used as features in different Random Forest classification models that are compared against each other. Our results show that the most relevant website characteristics are textual content, the use of English language, the number of subpages and the amount of characters on a website. In our main analysis, models using all website characteristics jointly yield AUC values of up to 0.75 and increase accuracy scores by up to 18 percentage points compared to a baseline prediction based on the sample mean. Moreover, predictions with website characteristics significantly differ from baseline predictions according to a McNemar test. Results also indicate a better performance for the prediction of product innovators and firms with innovation expenditures than for the prediction of process innovators. [ABSTRACT FROM AUTHOR]

Background and objectives: Diagnosis of Pulmonary Rifampicin Resistant Tuberculosis (RR-TB) with the Drug-Susceptibility Test (DST) is costly and time-consuming. Furthermore, GeneXpert for rapid diagnosis is not widely available in Indonesia. This study aims to develop and evaluate the CUHAS-ROBUST model performance, an artificial-intelligence-based RR-TB screening tool. Methods: A cross-sectional study involved suspected all type of RR-TB patients with complete sputum Lowenstein Jensen DST (reference) and 19 clinical, laboratory, and radiology parameter results, retrieved from medical records in hospitals under the Faculty of Medicine, Hasanuddin University Indonesia, from January 2015-December 2019. The Artificial Neural Network (ANN) models were built along with other classifiers. The model was tested on participants recruited from January 2020-October 2020 and deployed into CUHAS-ROBUST (index test) application. Sensitivity, specificity, and accuracy were obtained for assessment. Results: A total of 487 participants (32 Multidrug-Resistant/MDR 57 RR-TB, 398 drug-sensitive) were recruited for model building and 157 participants (23 MDR and 21 RR) in prospective testing. The ANN full model yields the highest values of accuracy (88% (95% CI 85–91)), and sensitivity (84% (95% CI 76–89)) compare to other models that show sensitivity below 80% (Logistic Regression 32%, Decision Tree 44%, Random Forest 25%, Extreme Gradient Boost 25%). However, this ANN has lower specificity among other models (90% (95% CI 86–93)) where Logistic Regression demonstrates the highest (99% (95% CI 97–99)). This ANN model was selected for the CUHAS-ROBUST application, although still lower than the sensitivity of global GeneXpert results (87.5%). Conclusion: The ANN-CUHAS ROBUST outperforms other AI classifiers model in detecting all type of RR-TB, and by deploying into the application, the health staff can utilize the tool for screening purposes particularly at the primary care level where the GeneXpert examination is not available. Trial registration: NCT04208789. [ABSTRACT FROM AUTHOR]

Taxi order demand prediction is of tremendous importance for continuous upgrading of an intelligent transportation system to realise city-scale and personalised services. An accurate short-term taxi demand prediction model in both spatial and temporal relations can assist a city pre-allocate its resources and facilitate city-scale taxi operation management in a megacity. To address problems similar to the above, in this study, we proposed a multi-zone order demand prediction model to predict short-term taxi order demand in different zones at city-scale. A two-step methodology was developed, including order zone division and multi-zone order prediction. For the zone division step, the K-means++ spatial clustering algorithm was used, and its parameter k was estimated by the between–within proportion index. For the prediction step, six methods (backpropagation neural network, support vector regression, random forest, average fusion-based method, weighted fusion-based method, and k-nearest neighbour fusion-based method) were used for comparison. To demonstrate the performance, three multi-zone weighted accuracy indictors were proposed to evaluate the order prediction ability at city-scale. These models were implemented and validated on real-world taxi order demand data from a three-month consecutive collection in Shenzhen, China. Experiment on the city-scale taxi demand data demonstrated the superior prediction performance of the multi-zone order demand prediction model with the k-nearest neighbour fusion-based method based on the proposed accuracy indicator. [ABSTRACT FROM AUTHOR]

Biological invasions have created detrimental impacts in freshwater ecosystems. As non-native freshwater species include economically beneficial, but also harmful, species, trait-based risk assessments can be used to identify and prevent the import of potentially invasive species. Freshwater fishes are one of the most evaluated freshwater taxa to date. However, such assessments have mostly been done in sub-temperate to temperate regions, with a general lack of such research in the tropics. In view of this knowledge gap, this study aims to determine if a different set of traits are associated with successful establishment of non-native fishes within the tropics. In tropical Southeast Asia, Singapore represents a suitable model site to perform an invasive species trait-based risk assessment for the tropical region given its susceptibility to the introduction and establishment of non-native freshwater fishes and lack of stringent fish import regulation. A quantitative trait-based risk assessment was performed using random forest to determine the relative importance of species attributes associated with the successful establishment of introduced freshwater fishes in Singapore. Species having a match in climate, prior invasion success, lower absolute fecundity, higher trophic level, and involvement in the aquarium trade were found to have higher establishment likelihood (as opposed to native distributional range and maximum size being among the commonly identified predictors in subtropical/temperate trait-based risk assessments). To minimize invasive risk, incoming freshwater fishes could be screened in future for such traits, allowing lists of prohibited or regulated species to be updated. The findings could also potentially benefit the development of invasive species action plans and inform management decisions in the Southeast Asian region. Considering a geographical bias in terms of having relatively less documentation of biological invasions in the tropics, particularly Asia, this study highlights the need to perform more of such risk assessments in other parts of the tropics. [ABSTRACT FROM AUTHOR]

Identification of success factors in wrestling as well as establishing their hierarchy are crucial from a cognitive and practical standpoint. It may provide a lot of practical recommendations related to wrestling-specific training. The aim of this study was to identify and establish the hierarchy of success factors in wrestling regardless of a fighting style and weight class. This study included 168 elite male freestyle and Greco-Roman wrestlers. They were divided into two groups: athletes who won medals (successful wrestlers) in high-rank competitions (Polish Championships or higher) and those who did not win any medals (less successful wrestlers) in those competitions. The following elements were assessed: anthropological measurements, body composition, dynamic strength, strength endurance, agility, special endurance, wrestling-specific fitness, response time, technical wrestling skills and anaerobic capacity. For initial data analysis, one-way ANOVA (α = 0.005) was used. Random Forests classifier was employed to identify success factors and to determine the importance of each of these factors in terms of sports performance. Seven key success factors were identified: anaerobic power, strength endurance, response time, special endurance, wrestling-specific fitness and technical wrestling skills performed in a horizontal position. Random Forests turned out to be an effective method of modelling success in wrestling (compared to SVM and KNN, which were also used in the study). These findings suggest that wrestling-specific training can be effectively monitored by controlling several vital indicators of athletes' preparedness: anaerobic power, strength endurance, response time, special endurance, wrestling-specific fitness and technical wrestling skills (the performance of reverse waistlock from a standing position and trunk grip gut wrench assessed by experts). [ABSTRACT FROM AUTHOR]

This study aimed to investigate the possible incidence of visual light perceptions (VLPs) during radiation therapy (RT). We analyzed whether VLPs could be affected by differences in the radiation energy, prescription doses, age, sex, or RT locations, and whether all VLPs were caused by radiation. From November 2016 to August 2018, a total of 101 patients who underwent head-and-neck or brain RT were screened. After receiving RT, questionnaires were completed, and the subjects were interviewed. Random forests (RF), a tree-based machine learning algorithm, and logistic regression (LR) analyses were compared by the area under the curve (AUC), and the algorithm that achieved the highest AUC was selected. The dataset sample was based on treatment with non-human units, and a total of 293 treatment fields from 78 patients were analyzed. VLPs were detected only in 122 of the 293 exposure portals (40.16%). The dataset was randomly divided into 80% and 20% as the training set and test set, respectively. In the test set, RF achieved an AUC of 0.888, whereas LR achieved an AUC of 0.773. In this study, the retina fraction dose was the most important continuous variable and had a positive effect on VLP. Age was the most important categorical variable. In conclusion, the visual light perception phenomenon by the human body during RT is induced by radiation rather than being a self-suggested hallucination or induced by phosphenes. [ABSTRACT FROM AUTHOR]

Mental fatigue is a common problem associated with neurological disorders. Until now, there has not been a method to assess mental fatigue on a continuous scale. Camera-based eye-typing is commonly used for communication by people with severe neurological disorders. We designed a working memory-based eye-typing experiment with 18 healthy participants, and obtained eye-tracking and typing performance data in addition to their subjective scores on perceived effort for every sentence typed and mental fatigue, to create a model of mental fatigue for eye-typing. The features of the model were the eye-based blink frequency, eye height and baseline-related pupil diameter. We predicted subjective ratings of mental fatigue on a six-point Likert scale, using random forest regression, with 22% lower mean absolute error than using simulations. When additionally including task difficulty (i.e. the difficulty of the sentences typed) as a feature, the variance explained by the model increased by 9%. This indicates that task difficulty plays an important role in modelling mental fatigue. The results demonstrate the feasibility of objective and non-intrusive measurement of fatigue on a continuous scale. [ABSTRACT FROM AUTHOR]

We present gridded 8 km-resolution data products of the estimated stem density, basal area, and biomass of tree taxa at Euro-American settlement of the midwestern United States during the middle to late 19th century for the states of Minnesota, Wisconsin, Michigan, Illinois, and Indiana. The data come from settlement-era Public Land Survey (PLS) data (ca. 0.8-km resolution) of trees recorded by land surveyors. The surveyor notes have been transcribed, cleaned, and processed to estimate stem density, basal area, and biomass at individual points. The point-level data are aggregated within 8 km grid cells and smoothed using a generalized additive statistical model that accounts for zero-inflated continuous data and provides approximate Bayesian uncertainty estimates. The statistical modeling smooths out sharp spatial features (likely arising from statistical noise) within areas smaller than about 200 km2. Based on this modeling, presettlement Midwestern landscapes supported multiple dominant species, vegetation types, forest types, and ecological formations. The prairies, oak savannas, and forests each had distinctive structures and spatial distributions across the domain. Forest structure varied from savanna (averaging 27 Mg/ha biomass) to northern hardwood (104 Mg/ha) and mesic southern forests (211 Mg/ha). The presettlement forests were neither unbroken and massively-statured nor dominated by young forests constantly structured by broad-scale disturbances such as fire, drought, insect outbreaks, or hurricanes. Most forests were structurally between modern second growth and old growth. We expect the data product to be useful as a baseline for investigating how forest ecosystems have changed in response to the last several centuries of climate change and intensive Euro-American land use and as a calibration dataset for paleoecological proxy-based reconstructions of forest composition and structure for earlier time periods. The data products (including raw and smoothed estimates at the 8-km scale) are available at the LTER Network Data Portal as version 1.0. [ABSTRACT FROM AUTHOR]

Reference evapotranspiration (ETo) is a fundamental parameter for hydrological studies and irrigation management. The Penman-Monteith method is the standard to estimate ETo and requires several meteorological elements. In developing countries, the number of weather stations is insufficient. Thus, free products of remote sensing with evapotranspiration information must be used for this purpose. In this context, the objective of this study was to estimate monthly ETo from potential evapotranspiration (PET) made available by MOD16 product. In this study, the monthly ETo estimated by Penman-Monteith method was considered as the standard. For this, data from 265 weather station of the National Institute of Meteorology (INMET), spread all over the Brazilian territory, were acquired for the period from 2000 to 2014 (15 years). For these months, monthly PET values from MOD16 product for all Brazil were also downloaded. By using machine learning algorithms and information from WorldClim as covariates, ETo was estimated through images from the MOD16 product. To perform the modeling of ETo, eight regression algorithms were tested: multiple linear regression; random forest; cubist; partial least squares; principal components regression; adaptive forward-backward greedy; generalized boosted regression and generalized linear model by likelihood-based boosting. Data from 2000 to 2012 (13 years) were used for training and data of 2013 and 2014 (2 years) were used to test the models. The PET made available by the MOD16 product showed higher values than those of ETo for different periods and climatic regions of Brazil. However, the MOD16 product showed good correlation with ETo, indicating that it can be used in ETo estimation. All models of machine learning were effective in improving the performance of the metrics evaluated. Cubist was the model that presented the best metrics for r2 (0.91), NSE (0.90) and nRMSE (8.54%) and should be preferred for ETo prediction. MOD16 product is recommended to be used to predict monthly ETo, which opens possibilities for its use in several other studies. [ABSTRACT FROM AUTHOR]

Risk assessment of in-hospital mortality of patients at the time of hospitalization is necessary for determining the scale of required medical resources for the patient depending on the patient's severity. Because recent machine learning application in the clinical area has been shown to enhance prediction ability, applying this technique to this issue can lead to an accurate prediction model for in-hospital mortality prediction. In this study, we aimed to generate an accurate prediction model of in-hospital mortality using machine learning techniques. Patients 18 years of age or older admitted to the University of Tokyo Hospital between January 1, 2009 and December 26, 2017 were used in this study. The data were divided into a training/validation data set (n = 119,160) and a test data set (n = 33,970) according to the time of admission. The prediction target of the model was the in-hospital mortality within 14 days. To generate the prediction model, 25 variables (age, sex, 21 laboratory test items, length of stay, and mortality) were used to predict in-hospital mortality. Logistic regression, random forests, multilayer perceptron, and gradient boost decision trees were performed to generate the prediction models. To evaluate the prediction capability of the model, the model was tested using a test data set. Mean probabilities obtained from trained models with five-fold cross-validation were used to calculate the area under the receiver operating characteristic (AUROC) curve. In a test stage using the test data set, prediction models of in-hospital mortality within 14 days showed AUROC values of 0.936, 0.942, 0.942, and 0.938 for logistic regression, random forests, multilayer perceptron, and gradient boosting decision trees, respectively. Machine learning-based prediction of short-term in-hospital mortality using admission laboratory data showed outstanding prediction capability and, therefore, has the potential to be useful for the risk assessment of patients at the time of hospitalization. [ABSTRACT FROM AUTHOR]

Purpose: The goal of this study is to construct a mortality prediction model using the XGBoot (eXtreme Gradient Boosting) decision tree model for AKI (acute kidney injury) patients in the ICU (intensive care unit), and to compare its performance with that of three other machine learning models. Methods: We used the eICU Collaborative Research Database (eICU-CRD) for model development and performance comparison. The prediction performance of the XGBoot model was compared with the other three machine learning models. These models included LR (logistic regression), SVM (support vector machines), and RF (random forest). In the model comparison, the AUROC (area under receiver operating curve), accuracy, precision, recall, and F1 score were used to evaluate the predictive performance of each model. Results: A total of 7548 AKI patients were analyzed in this study. The overall in-hospital mortality of AKI patients was 16.35%. The best performing algorithm in this study was XGBoost with the highest AUROC (0.796, p < 0.01), F1(0.922, p < 0.01) and accuracy (0.860). The precision (0.860) and recall (0.994) of the XGBoost model rank second among the four models. Conclusion: XGBoot model had obvious advantages of performance compared to the other machine learning models. This will be helpful for risk identification and early intervention for AKI patients at risk of death. [ABSTRACT FROM AUTHOR]

Computational methods have been widely used in drug design. The recent developments in machine learning techniques and the ever-growing chemical and biological databases are fertile ground for discoveries in this area. In this study, we evaluated the performance of Deep Learning models in comparison to Random Forest, and Support Vector Regression for predicting the biological activity (pIC50) of ALK-5 inhibitors as candidates to treat cancer. The generalization power of the models was assessed by internal and external validation procedures. A deep neural network model obtained the best performance in this comparative study, achieving a coefficient of determination of 0.658 on the external validation set with mean square error and mean absolute error of 0.373 and 0.450, respectively. Additionally, the relevance of the chemical descriptors for the prediction of biological activity was estimated using Permutation Importance. We can conclude that the forecast model obtained by the deep neural network is suitable for the problem and can be employed to predict the biological activity of new ALK-5 inhibitors. [ABSTRACT FROM AUTHOR]

In this paper we focus on a critical component of the city: its building stock, which holds much of its socio-economic activities. In our case, the lack of a comprehensive database about their features and its limitation to a surveyed subset lead us to adopt data-driven techniques to extend our knowledge to the near-city-scale. Neural networks and random forests are applied to identify the buildings' number of floors and construction periods' dependencies on a set of shape features: area, perimeter, and height along with the annual electricity consumption, relying a surveyed data in the city of Beirut. The predicted results are then compared with established scaling laws of urban forms, which constitutes a further consistency check and validation of our workflow. [ABSTRACT FROM AUTHOR]

Background: Compared to other studies, the injury monitoring of Chinese children and adolescents has captured a low level of intentional injuries on account of self-harm/suicide and violent attacks. Intentional injuries in children and adolescents have not been apparent from the data. It is possible that there has been a misclassification of existing intentional injuries, and there is a lack of research literature on the misclassification of intentional injuries. This study aimed to discuss the feasibility of discriminating the intention of injury based on Machine Learning (ML) modelling and provided ideas for understanding whether there was a misclassification of intentional injuries. Methods: Information entropy was used to determine the correlation between variables and the intention of injury, and Naive Bayes (NB), Decision Tree (DT), Random Forest (RF), Adaboost algorithms and Deep Neural Networks (DNN) were used to create an intention of injury discrimination model. The models were compared by comprehensively testing the discrimination effect to determine stability and consistency. Results: For the area under the ROC curve with different intentions of injuries, the NB model was 0.891, 0.880, and 0.897, respectively; the DT model was 0.870, 0.803, and 0.871, respectively; the RF model was 0.850, 0.809, and 0.845, respectively; the Adaboost model was 0.914, 0.846, and 0.914, respectively; the DNN model was 0.927, 0.835, and 0.934, respectively. In a comprehensive comparison of the five models, DNN and Adaboost models had higher values for the determination of the intention of injury. A discrimination of cases with unclear intentions of injury showed that on average, unintentional injuries, violent attacks, and self-harm/suicides accounted for 86.57%, 6.81%, and 6.62%, respectively. Conclusion: It was feasible to use the ML algorithm to determine the injury intention of children and adolescents. The research suggested that the DNN and Adaboost models had higher values for the determination of the intention of injury. This study could build a foundation for transforming the model into a tool for rapid diagnosis and excavating potential intentional injuries of children and adolescents by widely collecting the influencing factors, extracting the influence variables characteristically, reducing the complexity and improving the performance of the models in the future. [ABSTRACT FROM AUTHOR]

This study aimed to evaluate the validity and precision of the International Physical Activity Questionnaire (IPAQ) for climacteric women using computational intelligence techniques. The instrument was applied to 873 women aged between 40 and 65 years. Considering the proposal to regroup the set of data related to the level of physical activity of climacteric women using the IPAQ, we used 2 algorithms: Kohonen and k-means, and, to evaluate the validity of these clusters, 3 indexes were used: Silhouette, PBM and Dunn. The questionnaire was tested for validity (factor analysis) and precision (Cronbach's alpha). The Random Forests technique was used to assess the importance of the variables that make up the IPAQ. To classify these variables, we used 3 algorithms: Suport Vector Machine, Artificial Neural Network and Decision Tree. The results of the tests to evaluate the clusters suggested that what is recommended for IPAQ, when applied to climacteric women, is to categorize the results into two groups. The factor analysis resulted in three factors, with factor 1 being composed of variables 3 to 6; factor 2 for variables 7 and 8; and factor 3 for variables 1 and 2. Regarding the reliability estimate, the results of the standardized Cronbach's alpha test showed values between 0.63 to 0.85, being considered acceptable for the construction of the construct. In the test of importance of the variables that make up the instrument, the results showed that variables 1 and 8 presented a lesser degree of importance and by the analysis of Accuracy, Recall, Precision and area under the ROC curve, there was no variation when the results were analyzed with all IPAQ variables but variables 1 and 8. Through this analysis, we concluded that the IPAQ, short version, has adequate measurement properties for the investigated population. [ABSTRACT FROM AUTHOR]

Agent-based modeling of artificial societies allows for the validation and analysis of human-interpretable, causal explanations of human behavior that generate society-scale phenomena. However, parameter calibration is insufficient to conduct data-driven explorations that are adequate in evaluating the importance of causal factors that constitute agent rules that match real-world individual-scale generative behaviors. We introduce evolutionary model discovery, a framework that combines genetic programming and random forest regression to evaluate the importance of a set of causal factors hypothesized to affect the individual's decision-making process. With evolutionary model discovery, we investigated the farm plot seeking behavior of the Ancestral Pueblo of the Long House Valley simulated in the Artificial Anasazi model. We evaluated the importance of causal factors unconsidered in the original model, which we hypothesized to have affected the decision-making process. Our findings, concur with other archaeological studies on the Ancestral Pueblo communities during the Pueblo II period, which indicate the existence of cross-village polities, hierarchical organization, and dependence on the viability of the agricultural niche. Contrary to the original Artificial Anasazi model, where closeness was the sole factor driving farm plot selection, selection of higher quality land, distancing from failed farm plots, and desire for social presence are found to be more important. Finally, models updated with farm selection strategies designed by incorporating these insights showed significant improvements in accuracy and robustness over the original Artificial Anasazi model. [ABSTRACT FROM AUTHOR]

Background: A powerful risk model allows clinicians, at the bedside, to ensure the early identification of and decision-making for patients showing signs of developing physiological instability during treatment. The aim of this study was to enhance the identification of patients at risk for deterioration through an accurate model using electrolyte, metabolite, and acid-base parameters near the end of patients' intensive care unit (ICU) stays. Methods: This retrospective study included 5157 adult patients during the last 72 hours of their ICU stays. The patients from the MIMIC-III database who had serum lactate, pH, bicarbonate, potassium, calcium, glucose, chloride, and sodium values available, along with the times at which those data were recorded, were selected. Survivor data from the last 24 hours before discharge and four sets of nonsurvivor data from 48–72, 24–48, 8–24, and 0–8 hours before death were analyzed. Deep learning (DL), random forest (RF) and generalized linear model (GLM) analyses were applied for model construction and compared in terms of performance according to the area under the receiver operating characteristic curve (AUC). A DL backcasting approach was used to assess predictors of death vs. discharge up to 72 hours in advance. Results: The DL, RF and GLM models achieved the highest performance for nonsurvivors 0–8 hours before death versus survivors compared with nonsurvivors 8–24, 24–48 and 48–72 hours before death versus survivors. The DL assessment outperformed the RF and GLM assessments and achieved discrimination, with an AUC of 0.982, specificity of 0.947, and sensitivity of 0.935. The DL backcasting approach achieved discrimination with an AUC of 0.898 compared with the DL native model of nonsurvivors from 8–24 hours before death versus survivors with an AUC of 0.894. The DL backcasting approach achieved discrimination with an AUC of 0.871 compared with the DL native model of nonsurvivors from 48–72 hours before death versus survivors with an AUC of 0.846. Conclusions: The DL backcasting approach could be used to simultaneously monitor changes in the electrolyte, metabolite, and acid-base parameters of patients who develop physiological instability during ICU treatment and predict the risk of death over a period of hours to days. [ABSTRACT FROM AUTHOR]

Dietary nutrients have significant effects on the risk of cardiovascular diseases. However, the results were not uniform across different countries. The study aims to determine the relative importance of dietary nutrients associated with coronary artery disease (CAD) among the Nepalese population. A hospital-based matched case-control study was carried out at Shahid Gangalal National Heart Center in Nepal. In the present study, patients with more than seventy percent stenosis in any main coronary artery branch in angiography were defined as cases, while those presenting normal coronary angiography or negative for stressed exercise test were considered controls. Dietary intakes of 612 respondents over the past 12 months were evaluated using a semi-quantitative customized food frequency questionnaire. In conditional regression model, the daily average dietary intake of β-carotene (OR: 0.54; 95%CI: 0.34, 0.87), and vitamin C (OR: 0.96; 95%CI: 0.93, 0.99) were inversely, whereas dietary carbohydrate (OR: 1.16; 95%CI: 1.1, 1.24), total fat/oil (OR: 1.47; 95%CI: 1.27, 1.69), saturated fatty acid (SFA) (OR: 1.2; 95%CI: 1.11, 1.3), cholesterol (OR: 1.01; 95%CI: 1.001, 1.014), and iron intakes (OR: 1.11; 95%CI: 1.03, 1.19) were positively linked with CAD. Moreover, in random forest analysis, the daily average dietary intakes of SFA, vitamin A, total fat/oil, β-carotene, and cholesterol were among the top five nutrients (out of 12 nutrients variables) of relative importance associated with CAD. The nutrients of relative importance imply a reasonable preventive measure in public health nutrients specific intervention to prevent CAD in a resource-poor country like Nepal. The findings are at best suggestive of a possible relationship between these nutrients and the development of CAD, but prospective cohort studies and randomized control trials will need to be performed in the Nepalese population. [ABSTRACT FROM AUTHOR]

Evaluative research of technological officiating aids in sports predominantly focuses on the respective technology and the impact on decision accuracy, whereas the impact on stakeholders is neglected. Therefore, the aim of this study was to investigate the immediate impact of the recently introduced Video Assistant Referee, often referred to as VAR, on the sentiment of fans of the English Premier League. We analyzed the content of 643,251 tweets from 129 games, including 94 VAR incidents, using a new variation of a gradient boosting approach to train two tree-based classifiers for text corpora: one classifier to identify tweets related to the VAR and another one to rate a tweet's sentiment. The results of 10-fold cross-validations showed that our approach, for which we only took a small share of all features to grow each tree, performed better than common approaches (naïve Bayes, support vector machines, random forest and traditional gradient tree boosting) used by other studies for both classification problems. Regarding the impact of the VAR on fans, we found that the average sentiment of tweets related to this technological officiating aid was significantly lower compared to other tweets (-0.64 vs. 0.08; t = 45.5, p <.001). Further, by tracking the mean sentiment of all tweets chronologically for each game, we could display that there is a significant drop of sentiment for tweets posted in the periods after an incident compared to the periods before. A plunge that persisted for 20 minutes on average. Summed up, our results provide evidence that the VAR effects predominantly expressions of negative sentiment on Twitter. This is in line with the results found in previous, questionnaire-based, studies for other technological officiating aids and also consistent with the psychological principle of loss aversion. [ABSTRACT FROM AUTHOR]

Persons who inject drugs (PWID) are at increased risk for overdose death (ODD), infections with HIV, hepatitis B (HBV) and hepatitis C virus (HCV), and noninfectious health conditions. Spatiotemporal identification of PWID communities is essential for developing efficient and cost-effective public health interventions for reducing morbidity and mortality associated with injection-drug use (IDU). Reported ODDs are a strong indicator of the extent of IDU in different geographic regions. However, ODD quantification can take time, with delays in ODD reporting occurring due to a range of factors including death investigation and drug testing. This delayed ODD reporting may affect efficient early interventions for infectious diseases. We present a novel model, Dynamic Overdose Vulnerability Estimator (DOVE), for assessment and spatiotemporal mapping of ODDs in different U.S. jurisdictions. Using Google® Web-search volumes (i.e., the fraction of all searches that include certain words), we identified a strong association between the reported ODD rates and drug-related search terms for 2004–2017. A machine learning model (Extremely Random Forest) was developed to produce yearly ODD estimates at state and county levels, as well as monthly estimates at state level. Regarding the total number of ODDs per year, DOVE's error was only 3.52% (Median Absolute Error, MAE) in the United States for 2005–2017. DOVE estimated 66,463 ODDs out of the reported 70,237 (94.48%) during 2017. For that year, the MAE of the individual ODD rates was 4.43%, 7.34%, and 12.75% among yearly estimates for states, yearly estimates for counties, and monthly estimates for states, respectively. These results indicate suitability of the DOVE ODD estimates for dynamic IDU assessment in most states, which may alert for possible increased morbidity and mortality associated with IDU. ODD estimates produced by DOVE offer an opportunity for a spatiotemporal ODD mapping. Timely identification of potential mortality trends among PWID might assist in developing efficient ODD prevention and HBV, HCV, and HIV infection elimination programs by targeting public health interventions to the most vulnerable PWID communities. [ABSTRACT FROM AUTHOR]