Search

Accurately predicting spectra for heavy elements, often open-shell systems, is a significant challenge typically addressed using a single cluster approach with a fixed coordination number. Developing a realistic model that accounts for temperature effects, variable coordination numbers, and interprets experimental data is even more demanding due to the strong solute–solvent interactions present in solutions of heavy metal cations. This study addresses these challenges by combining multiple methodologies to accurately predict realistic spectra for highly charged metal cations in aqueous media, with a focus on the electronic absorption spectrum of Ce3+ in water. Utilizing highly correlated relativistic quantum mechanical (QM) wavefunctions and structures from molecular dynamics (MD) simulations, we show that the convolution of individual vertical transitions yields excellent agreement with experimental results without the introduction of empirical broadening. Good results are obtained for both the normalized spectrum and that of absolute intensity. The study incorporates a statistical machine learning algorithm, Gaussian Mixture Models-Nuclear Ensemble Approach (GMM-NEA), to convolute individual spectra. The microscopic distribution provided by MD simulations allows us to examine the contributions of the octa- and ennea-hydrate of Ce3+ in water to the final spectrum. In addition, the temperature dependence of the spectrum is theoretically captured by observing the changing population of these hydrate forms with temperature. We also explore an alternative method for obtaining statistically representative structures in a less demanding manner than MD simulations, derived from QM Wigner distributions. The combination of Wigner-sampling and GMM-NEA broadening shows promise for wide application in spectroscopic analysis and predictions, offering a computationally efficient alternative to traditional methods. [ABSTRACT FROM AUTHOR]

The ever-happening disruptive events interrupt the operationalisation of manufacturing organisations resulting in stalling the production flow and depleting societies with products. Advancements in cutting-edge technologies, viz. blockchain, artificial intelligence, virtual reality, digital twin, etc. have attracted the practitioners' attention to overcome such saddled conditions. This study attempts to explore the role of artificial intelligence (AI) in building the resilience of production function at manufacturing organisations during a COVID-19 pandemic. In this regard, a decision support system comprising an integrated voting analytical hierarchy process (VAHP) and Bayesian network (BN) method is developed. Initially, through a comprehensive literature review, the critical success factors (CSFs) for implementing AI are determined. Further, using a multi-criteria decision-making (MCDM) based VAHP, CSFs are prioritised to determine the prominent ones. Finally, the machine learning based BN method is adopted to predict and understand the influential CSFs that help achieve the highest production resilience. The present research is one of the early attempts to know the essence of AI and bridge the interplay between AI and production resilience during COVID-19. This study can support academicians, practitioners, and decision-makers in assessing the AI adoption in manufacturing organisations and evaluate the impact of different CSFs of AI on production resilience. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of stochastic jointly scheduling of resumable jobs and preventive maintenance on a single machine, subject to random breakdowns, to minimise the earliness-tardiness cost. The main objective is to investigate using trending machine learning-based methods compared to stochastic optimisation approaches. We propose two different methods from both fields as we solve the same problem firstly with a stochastic dynamic programming model in an approximation way, then with an attention-based deep reinforcement learning model. We conduct a detailed experimental study according to solution quality, run time, and robustness to analyse their performances compared to those of an existing approach in the literature as a baseline. Both algorithms outperform the baseline. Moreover, the machine learning-based algorithm outperforms the stochastic dynamic programming-based heuristic as we report up to 30.5% saving in total cost, a reduction of computational time from 67 min to less than $ 1s $ 1 s on big instances, and a better robustness. These facts highlight clearly its potential for solving such problems. [ABSTRACT FROM AUTHOR]

The bridge natural frequency changes caused by structural damages are often masked by the variations of environmental factors (especially the temperature), thus the ability to detect early damage is usually weakened. In order to tolerate the environmental effects, this paper proposes to normalize the natural frequency data based on probabilistic machine learning for the early damage detection of bridge structures. First, the probabilistic nonlinear relationships between multi-order natural frequencies in the intact state are modeled by the Gaussian process regression or the relevance vector machine. Second, the normalized frequency residuals for each order of frequency are individually calculated through their conditional means and variances estimated from the built probabilistic nonlinear relationship models. Third, the lifting residual vectors corresponding to the normalized frequency residuals are constructed according to the statistical localization approach. And finally, the Mahalanobis squared distance of the residual vector is defined as the damage indicator, afterwards the corresponding damage threshold is determined by the kernel density estimation technique. The long-term natural frequency data from a numerical simulation and an actual bridge are used to verify the effectiveness of the proposed method. The results demonstrate that the normalized frequency residuals are scarcely affected by the environmental variations, while remaining high sensitivity to the structural damages; compared to the normalized frequency residual, the damage indicator defined from the lifting residual vector is more suitable for detecting the early damage of bridge structures. [ABSTRACT FROM AUTHOR]

AbstractU-statistics play central roles in many statistical learning tools but face the haunting issue of scalability. Despite extensive research on accelerating computation by U-statistic reduction, existing results almost exclusively focused on power analysis. Little work addresses risk control accuracy, which requires distinct and much more challenging techniques. In this paper, we establish the first statistical inference procedure with provably higher-order accurate risk control for incomplete U-statistics. The sharpness of our new result enables us to reveal how risk control accuracy also trades off with speed, for the first time in literature, which complements the well-known variance-speed trade-off. Our general framework converts the challenging and case-by-case analysis for many different designs into a surprisingly principled and routine computation. We conducted comprehensive numerical studies and observed results that validate our theory’s sharpness. Our method also demonstrates effectiveness on real-world data applications. [ABSTRACT FROM AUTHOR]

Scenario optimization is an approach to data-driven decision-making that has been introduced some fifteen years ago and has ever since then grown fast. Its most remarkable feature is that it blends the heuristic nature of data-driven methods with a rigorous theory that allows one to gain factual, reliable, insight in the solution. The usability of the scenario theory, however, has been restrained thus far by the obstacle that most results are standing on the assumption of convexity. With this paper, we aim to free the theory from this limitation. Specifically, we focus on the body of results that are known under the name of "wait-and-judge" and show that its fundamental achievements maintain their validity in a non-convex setup. While optimization is a major center of attention, this paper travels beyond it and into data-driven decision making. Adopting such a broad framework opens the door to building a new theory of truly vast applicability. [ABSTRACT FROM AUTHOR]

Human factors contribute in some way to almost 93% of road crashes. Because the unavailability of good pre-crash data, the contribution of human factors to safety-critical events (SCEs) and the prediction of crashes using real-world data is lightly researched. This study provides predictive accuracy by harnessing unique real-world naturalistic driving study (NDS) data, which includes dynamic pre-crash information about driving behavior and performance. After cleaning and preprocessing, a final subsample (N = 9,237) was used and split into training and test samples. For consistent comparison of variables' importance in statistical and machine learning (ML) models, the dominance analysis uncovered the most important predictors used by the ordered Probit model. Next, three non-parametric supervised ML methods, because promising prediction performance and cost-effectiveness, including Naïve Bayes, K-Nearest Neighbors, and Gradient Boosting Tree (GBT) were used. The overall out-of-sample prediction accuracy for the ordered Probit model was 85.75% which was lower than all three ML methods. The GBT showed the highest (91.23%) out-of-sample prediction accuracy. The availability of pre-crash naturalistic data helps significantly improve the prediction accuracy of SCEs as cumulative importance for all available human factors in the GBT classifier was 94%. For practical applications, refer to the article. [ABSTRACT FROM AUTHOR]

Caregivers play an outsized role in shaping early life experiences and development, but we often lack mechanistic insight into how exactly caregiver behavior scaffolds the neurodevelopment of specific learning processes. Here, we capitalized on the fact that caregivers differ in how predictable their behavior is to ask if infants' early environmental input shapes their brains' later ability to learn about predictable information. As part of an ongoing longitudinal study in South Africa, we recorded naturalistic, dyadic interactions between 103 (46 females and 57 males) infants and their primary caregivers at 3–6 months of age, from which we calculated the predictability of caregivers' behavior, following caregiver vocalization and overall. When the same infants were 6–12‐months‐old they participated in an auditory statistical learning task during EEG. We found evidence of learning‐related change in infants' neural responses to predictable information during the statistical learning task. The magnitude of statistical learning‐related change in infants' EEG responses was associated with the predictability of their caregiver's vocalizations several months earlier, such that infants with more predictable caregiver vocalization patterns showed more evidence of statistical learning later in the first year of life. These results suggest that early experiences with caregiver predictability influence learning, providing support for the hypothesis that the neurodevelopment of core learning and memory systems is closely tied to infants' experiences during key developmental windows. [ABSTRACT FROM AUTHOR]

Although algorithms are imbued with a sense of objectivity and reliability, numerous high-profile incidents have demonstrated their fallibility. In response, many have called for algorithmic governance that mitigates their potential harms. Further, these incidents have inspired studies that consider algorithms as part of wider sociotechnical systems. In this article, we build on such work and focus on how the specific forms of algorithms may facilitate or constrain the ways in which they become embedded within these systems. More specifically, we suggest that (a) algorithms should be understood as models, with (b) divergent forms, and (c) associated representational qualities. We showcase this approach in three critical case studies of algorithmic models used in government: the SAFFIER II model that underpins the Netherlands government's spending, the Ofqual DCP A-Level grading algorithm that was used (and later abandoned) in lieu of actual secondary school exams in the United Kingdom, and the Risk Classification Model used by the Dutch Tax and Customs Administration to identify social benefit fraud. With the three case studies, we show how the divergent forms of algorithms have implications for their responsiveness and ultimately their solidification in – or dissolution from – socio-technical systems. [ABSTRACT FROM AUTHOR]

Language comprehension involves the grouping of words into larger multiword chunks. This is required to recode information into sparser representations to mitigate memory limitations and counteract forgetting. It has been suggested that electrophysiological processing time windows constrain the formation of these units. Specifically, the period of rhythmic neural activity (i.e., low-frequency neural oscillations) may set an upper limit of 2–3 sec. Here, we assess whether learning of new multiword chunks is also affected by this neural limit. We applied an auditory statistical learning paradigm of an artificial language while manipulating the duration of to-be-learned chunks. Participants listened to isochronous sequences of disyllabic pseudowords from which they could learn hidden three-word chunks based on transitional probabilities. We presented chunks of 1.95, 2.55, and 3.15 sec that were created by varying the pause interval between pseudowords. In a first behavioral experiment, we tested learning using an implicit target detection task. We found better learning for chunks of 2.55 sec as compared to longer durations in line with an upper limit of the proposed time constraint. In a second experiment, we recorded participants' electroencephalogram during the exposure phase to use frequency tagging as a neural index of statistical learning. Extending the behavioral findings, results show a significant decline in neural tracking for chunks exceeding 3 sec as compared to both shorter durations. Overall, we suggest that language learning is constrained by endogenous time constraints, possibly reflecting electrophysiological processing windows. [ABSTRACT FROM AUTHOR]

The human visual system is equipped to rapidly and implicitly learn and exploit the statistical regularities in our environment. Within visual search, contextual cueing demonstrates how implicit knowledge of scenes can improve search performance. This is commonly interpreted as spatial context in the scenes becoming predictive of the target location, which leads to a more efficient guidance of attention during search. However, what drives this enhanced guidance is unknown. First, it is under debate whether the entire scene (global context) or more local context drives this phenomenon. Second, it is unclear how exactly improved attentional guidance is enabled by target enhancement and distractor suppression. In the present magnetoencephalography experiment, we leveraged rapid invisible frequency tagging to answer these two outstanding questions. We found that the improved performance when searching implicitly familiar scenes was accompanied by a stronger neural representation of the target stimulus, at the cost specifically of those distractors directly surrounding the target. Crucially, this biasing of local attentional competition was behaviorally relevant when searching familiar scenes. Taken together, we conclude that implicitly learned spatial predictive context improves how we search our environment by sharpening the attentional field. [ABSTRACT FROM AUTHOR]

Contemporary geoengineering challenges grapple with the plateauing of both existing algorithms and their depth of insights, a phenomenon exacerbated by the scarcity of high‐fidelity data. Although existing solutions such as Monte‐Carlo method can generate abundant data, they are not sufficiently robust for ensuring the high fidelity of data. This study proposes a novel data augmentation framework that combines statistical and machine learning methods to generate high‐fidelity synthetic data, which closely align with field data in terms of the statistical and empirical attributes. The innovations of the proposed approach lie in the integration of Copulas theory for data generation, a developed geo‐regression anomaly detection (GRAD) for adjusting data attributes, and an evolutionary polynomial regression for data consistency enforcement. The multilayer perceptron (MLP) and a wide‐and‐deep (WaD) networks are applied to assess the effectiveness of high‐fidelity data augmentation using jet grouting data. The outcomes reveal the robustness of the synthetic data generation framework, achieving satisfactory fidelity in both empirical and statistical attributes. The proposed data augmentation improved the R2 and MAE achieved by MLP and WaD up to 28.37% under data fractions ranging from 0.2 to 1. MLP and WaD yielded comparable results in terms of accuracy and generalization ability across various augmented fractions. This indicates that the accuracy of synthetic data plays a pivotal role, suggesting improving data quality can be highly effective in boosting performance, regardless of the model complexity. This study contributes valuable insights to addressing the challenges of scare high‐fidelity data in geoengineering. [ABSTRACT FROM AUTHOR]

In the Namib Desert, fog is the only regular water input and, thus, is a crucial water source for its fauna and flora. Each year, between June and October, absorbing biomass burning aerosols (BBAs) overlie the stratocumulus clouds in the adjacent Southeast Atlantic. In some synoptic settings, this layer of BBAs reaches Namibia and its desert, where it interacts with coastal fog and low clouds (FLCs). In this study, a novel 15-year data set of geostationary satellite observations of FLC dissipation time in the Namib Desert is used, along with reanalysis data, to better understand the potential semi-direct effects of BBAs on FLC dissipation in the Namib Desert, i.e., through adjustments of atmospheric stability and thermodynamics via the interaction of aerosols with radiation. This is done by investigating both the time of day when FLCs dissolve and the synoptic-scale meteorology depending on BBA loading. It is found that FLC dissipation time is significantly later on high-BBA-loading days. BBAs are transported to the Namib along moist free-tropospheric air by a large-scale anticyclonic recirculation pattern. At the surface, the associated longwave heating strengthens a continental heat low, which modifies the circulation and boundary layer moisture along the coastline, complicating the attribution of BBA effects. During high-BBA days, the vertical profiles of the temporal development of air temperatures highlight contrasting daytime and nighttime processes modifying the local inversion. These processes are thought to be driven by greenhouse warming as a result of the moisture in the BBA plumes and BBA absorption (only during the daytime). A statistical learning framework is used to quantify meteorological and BBA influences on FLC dissipation time. The statistical model is able to reproduce the observed differences in FLC dissipation time between high- and low-BBA days and attributes these differences mainly to differences in circulation, boundary layer moisture and near-surface air temperature along the coastline. However, the model is prone to underfitting and is not able to reproduce the majority of the FLC dissipation variability. While the model does not suggest that BBA patterns are important for FLC dissipation, the findings show how the moist BBA plumes modify local thermodynamics, to which FLC dissipation is shown to be sensitive. The findings highlight the challenges of disentangling meteorological and aerosol effects on cloud development using observations and invite detailed modeling analyses of the underlying processes, for example, with large-eddy simulations. [ABSTRACT FROM AUTHOR]

Discovering and addressing unknown, including unanticipated, part-to-part variation sources is an important, yet challenging problem in manufacturing variation reduction. The state-of-art methods for solving this problem have focused solely on traditional mass manufacturing settings, in which abundant measurement data of parts with the same design are available. Applying these methods to custom manufacturing processes is problematic because the number of parts with the same design in custom manufacturing is often small. This paper proposes a new variation model that considers custom manufacturing parameters to aggregate measurement data across all custom parts. We also propose to estimate this model via a conditional autoencoder. The advantages of the proposed approach are demonstrated with a simulated toy-building brick example and a real cylindrical machining example. The approach successfully reveals unknown variation patterns even with a relatively small number of parts in these examples. Our approach is also generally applicable to any mainstream manufacturing processes that produce multiple part designs. [ABSTRACT FROM AUTHOR]

Introduction: Implicit statistical learning is, by definition, learning that occurs without conscious awareness. However, measures that putatively assess implicit statistical learning often require explicit reflection, for example, deciding if a sequence is 'grammatical' or 'ungrammatical'. By contrast, 'processing-based' tasks can measure learning without requiring conscious reflection, by measuring processes that are facilitated by implicit statistical learning. For example, when multiple stimuli consistently co-occur, it is efficient to 'chunk' them into a single cognitive unit, thus reducing working memory demands. Previous research has shown that when sequences of phonemes can be chunked into 'words', participants are better able to recall these sequences than random ones. Here, in two experiments, we investigated whether serial visual recall could be used to effectively measure the learning of a more complex artificial grammar that is designed to emulate the between-word relationships found in language. Methods: We adapted the design of a previous Artificial Grammar Learning (AGL) study to use a visual serial recall task, as well as more traditional reflection-based grammaticality judgement and sequence completion tasks. After exposure to "grammatical" sequences of visual symbols generated by the artificial grammar, the participants were presented with novel testing sequences. After a brief pause, participants were asked to recall the sequence by clicking on the visual symbols on the screen in order. Results: In both experiments, we found no evidence of artificial grammar learning in the Visual Serial Recall task. However, we did replicate previously reported learning effects in the reflection-based measures. Discussion: In light of the success of serial recall tasks in previous experiments, we discuss several methodological factors that influence the extent to which implicit statistical learning can be measured using these tasks. [ABSTRACT FROM AUTHOR]

Kawasaki Disease (KD) is a rare febrile illness affecting infants and young children, potentially leading to coronary artery complications and, in severe cases, mortality if untreated. However, KD is frequently misdiagnosed as a common fever in clinical settings, and the inherent data imbalance further complicates accurate prediction when using traditional machine learning and statistical methods. This paper introduces two advanced approaches to address these challenges, enhancing prediction accuracy and generalizability. The first approach proposes a stacking model termed the Disease Classifier (DC), specifically designed to recognize minority class samples within imbalanced datasets, thereby mitigating the bias commonly observed in traditional models toward the majority class. Secondly, we introduce a combined model, the Disease Classifier with CTGAN (CTGAN-DC), which integrates DC with Conditional Tabular Generative Adversarial Network (CTGAN) technology to improve data balance and predictive performance further. Utilizing CTGAN-based oversampling techniques, this model retains the original data characteristics of KD while expanding data diversity. This effectively balances positive and negative KD samples, significantly reducing model bias toward the majority class and enhancing both predictive accuracy and generalizability. Experimental evaluations indicate substantial performance gains, with the DC and CTGAN-DC models achieving notably higher predictive accuracy than individual machine learning models. Specifically, the DC model achieves sensitivity and specificity rates of 95%, while the CTGAN-DC model achieves 95% sensitivity and 97% specificity, demonstrating superior recognition capability. Furthermore, both models exhibit strong generalizability across diverse KD datasets, particularly the CTGAN-DC model, which surpasses the JAMA model with a 3% increase in sensitivity and a 95% improvement in generalization sensitivity and specificity, effectively resolving the model collapse issue observed in the JAMA model. In sum, the proposed DC and CTGAN-DC architectures demonstrate robust generalizability across multiple KD datasets from various healthcare institutions and significantly outperform other models, including XGBoost. These findings lay a solid foundation for advancing disease prediction in the context of imbalanced medical data. [ABSTRACT FROM AUTHOR]

Foehn winds, characterised by abrupt temperature increases and wind speed changes, significantly impact regions on the leeward side of mountain ranges, e.g., by spreading wildfires. Understanding how foehn occurrences change under climate change is crucial. As foehn is a meteorological phenomenon, its prevalence has to be inferred from meteorological measurements employing suitable classification schemes. Hence, this approach is typically limited to specific periods for which the necessary data are available. We present a novel approach for reconstructing historical foehn occurrences using a combination of unsupervised and supervised probabilistic statistical learning methods. We utilise in situ measurements (available for recent decades) to train an unsupervised learner (finite mixture model) for automatic foehn classification. These labelled data are then linked to reanalysis data (covering longer periods) using a supervised learner (lasso or boosting). This allows us to reconstruct past foehn probabilities based solely on reanalysis data. Applying this method to ERA5 reanalysis data for six stations across Switzerland and Austria achieves accurate hourly reconstructions of north and south foehn occurrence, respectively, dating back to 1940. This paves the way for investigating how seasonal foehn patterns have evolved over the past 83 years, providing valuable insights into climate change impacts on these critical wind events. [ABSTRACT FROM AUTHOR]

This study aims to propose an original approach to the interpretability of the explanatory variables (features) in the well-known expected goals (xG) model for shot analysis in football. To do this, a new original sample of 7801 shots from Italy's Serie A (1 binary outcome and 26 features) for the 2022/2023 and 2023/2024 seasons were used, in which 8 new features of various types were introduced, integrating event data, performance data, and tracking data. Specifically, the performance of 8 statistical and machine learning (algorithmic) classifiers was compared. The focus was on two key aspects related to the field of explainable Artificial Intelligence (xAI), ‘accuracy’ and ‘explainability’, assessed using some appropriate metrics. Considering the accuracy metrics, among the statistical classifiers Binary Regression (BR) with the cloglog link function is the most effective. In contrast, among the algorithmic classifiers, xGBoost has the best performance but is slightly lower than the BR-cloglog. Regarding explainability, the primary contribution to the xG consistently comes from a small set of variables across all classifiers. The most influential features are the proximity to the goal, the shooting angle, and the shooter's visual angle. [ABSTRACT FROM AUTHOR]

Background: Rheumatology has experienced notable changes in the last decades. New drugs, including biologic agents and Janus kinase (JAK) inhibitors, have blossomed. Concepts such as window of opportunity, arthralgia suspicious for progression, or difficult-to-treat rheumatoid arthritis (RA) have appeared; and new management approaches and strategies such as treat-to-target have become popular. Statistical learning methods, gene therapy, telemedicine, or precision medicine are other advancements that have gained relevance in the field. To better characterize the research landscape and advances in rheumatology, automatic and efficient approaches based on natural language processing (NLP) should be used. Objectives: The objective of this study is to use topic modeling (TM) techniques to uncover key topics and trends in rheumatology research conducted in the last 23 years. Design: Retrospective study. Methods: This study analyzed 96,004 abstracts published between 2000 and December 31, 2023, drawn from 34 specialized rheumatology journals obtained from PubMed. BERTopic, a novel TM approach that considers semantic relationships among words and their context, was used to uncover topics. Up to 30 different models were trained. Based on the number of topics, outliers, and topic coherence score, two of them were finally selected, and the topics were manually labeled by two rheumatologists. Word clouds and hierarchical clustering visualizations were computed. Finally, hot and cold trends were identified using linear regression models. Results: Abstracts were classified into 45 and 47 topics. The most frequent topics were RA, systemic lupus erythematosus, and osteoarthritis. Expected topics such as COVID-19 or JAK inhibitors were identified after conducting dynamic TM. Topics such as spinal surgery or bone fractures have gained relevance in recent years; however, antiphospholipid syndrome or septic arthritis have lost momentum. Conclusion: Our study utilized advanced NLP techniques to analyze the rheumatology research landscape and identify key themes and emerging trends. The results highlight the dynamic and varied nature of rheumatology research, illustrating how interest in certain topics has shifted over time. [ABSTRACT FROM AUTHOR]

Healthcare supply chain management's (HSCM) significance to economic and societal development is huge. In today's very competitive market, supply chains have seen significant changes in the last several years. There is a need for technology that can handle the increasing complexity of today's dynamic supply chain activities. Both machine learning (ML) and the quick dissemination of information have the potential to revolutionize the supply chain. ML has spawned a slew of useful supply chain applications in recent years, HSCM has received comparatively less attention. In this study, we applied three ML algorithms such as gradient boosting (GB), histogram gradient boosting (HGB), and cat boosting (CB) with data preprocessing tools to predict whether the medicines are delivered on time or not in the HSCM. The data preprocessing tools are used to manage datasets and increase the performance of ML algorithms. There are three methods of feature selection that are applied in this study such as Pearson correlation, chi-square test, and principal component analysis to select the best features to push in the ML algorithms. The main results show the CB is the best algorithm with the highest accuracy, precision, recall, and f1 score with values respectively. The three ML algorithms are compared with other ML algorithms to show the robustness of the applied ML algorithms. We made a sensitivity analysis to show the chaining in learning rate (LR) and compute the accuracy of the ML algorithms. We show the CB is not sensitive to values between 0.1 and 1. [ABSTRACT FROM AUTHOR]

The kernel method is a tool that converts data to a kernel space where operation can be performed. When converted to a high-dimensional feature space by using kernel functions, the data samples are more likely to be linearly separable. Traditional machine learning methods can be extended to the kernel space, such as the radial basis function (RBF) network. As a kernel-based method, support vector machine (SVM) is one of the most popular nonparametric classification methods, and is optimal in terms of computational learning theory. Based on statistical learning theory and the maximum margin principle, SVM attempts to determine an optimal hyperplane by addressing a quadratic programming (QP) problem. Using Vapnik–Chervonenkis dimension theory, SVM maximizes generalization performance by finding the widest classification margin within the feature space. In this paper, kernel machines and SVMs are systematically introduced. We first describe how to turn classical methods into kernel machines, and then give a literature review of existing kernel machines. We then introduce the SVM model, its principles, and various SVM training methods for classification, clustering, and regression. Related topics, including optimizing model architecture, are also discussed. We conclude by outlining future directions for kernel machines and SVMs. This article functions both as a state-of-the-art survey and a tutorial. [ABSTRACT FROM AUTHOR]

Mobility tasks like the Timed Up and Go test (TUG), cognitive TUG (cogTUG), and walking with turns provide insights into the impact of Parkinson's disease (PD) on motor control, balance, and cognitive function. We assess the test–retest reliability of these tasks in 262 PD participants and 50 controls by evaluating machine learning models based on wearable-sensor-derived measures and statistical metrics. This evaluation examines total duration, subtask duration, and other quantitative measures across two trials. We show that the diagnostic accuracy for distinguishing PD from controls decreases by a mean of 1.8% between the first and the second trial, suggesting that task repetition may not be necessary for accurate diagnosis. Although the total duration remains relatively consistent between trials (intraclass correlation coefficient (ICC) = 0.62 to 0.95), greater variability is seen in subtask duration and sensor-derived measures, reflected in machine learning performance and statistical differences. Our findings also show that this variability differs not only between controls and PD participants but also among groups with varying levels of PD severity, indicating the need to consider population characteristics. Relying solely on total task duration and conventional statistical metrics to gauge the reliability of mobility tasks may fail to reveal nuanced variations in movement. [ABSTRACT FROM AUTHOR]

Phishing attacks are the biggest cybersecurity threats in the digital world. Attackers exploit users by impersonating real, authentic websites to obtain sensitive information such as passwords and bank statements. One common technique in these attacks is using malicious URLs. These malicious URLs mimic legitimate URLs, misleading users into interacting with malicious websites. This practice, URL phishing, presents a big threat to internet security, emphasizing the need for advanced detection methods. So we aim to enhance phishing URL detection by using machine learning and deep learning models, leveraging a set of low-level URL features derived from n-gram analysis. In this paper, we present a method for detecting malicious URLs using statistical features extracted from n-grams. These n-grams are extracted from the hexadecimal representation of URLs. We employed 4 experiments in our paper. The first 3 experiments used machine learning with the statistical features extracted from these n-grams, and the fourth experiment used these grams directly with deep learning models to evaluate their effectiveness. Also, we used Explainable AI (XAI) to explore the extracted features and evaluate their importance and role in phishing detection. A key advantage of our method is its ability to reduce the number of features required and reduce the training time by using fewer features after applying XAI techniques. This stands in contrast to the previous study, which relies on high-level URL features and needs pre-processing and a high number of features (87 high-level URL-based features). So our technique only uses statistical features extracted from n-grams and the n-gram itself, without the need for any high-level features. Our method is evaluated across different n-gram lengths (2, 4, 6, and 8), aiming to optimize detection accuracy. We conducted four experiments in our study. In the first experiment, we focused on extracting and using 12 common statistical features like mean, median, etc. In the first experiment, the XGBoost model achieved the highest accuracy using 8-gram features with 82.41%. In the second experiment, we expanded the feature set and extracted an additional 13 features, so our feature count became 25. XGBoost in the second experiment achieved the highest accuracy with 86.40%. Accuracy improvement continued in the third experiment, we extracted an additional 16 features (character count features), and these features increased XGBoost accuracy to 88.15% in the third experiment. In the fourth experiment, we directly fed n-gram representations into deep learning models. The Convolutional Neural Network (CNN) model achieved the highest accuracy of 94.09% in experiment four. Also, we applied XAI techniques, SHapley Additive exPlanations (SHAP), and Local Interpretable Model-agnostic Explanations (LIME). Through the explanation provided by XAI methods, we were able to determine the most important features in our feature set, enabling a reduction in feature count. Using fewer features (4, 7, 10, 13, 15), we got good accuracy compared to the 41 features used in experiment three and reduced the models' training times and complexity. This research aimed to enhance phishing URL detection by using machine learning and deep learning models, leveraging a set of low-level URL features derived from n-gram analysis. Our findings show the importance of using minimal statistical features to identify malicious URLs. Notably, the use of CNN had a great advancement, achieving an accuracy rate of 94.09% with using n-grams of URLs, surpassing traditional machine learning models. This achievement not only validates the efficacy of deep learning models in complex pattern recognition tasks but also highlights the efficiency of our feature selection approach, which relies on a lower number of features and is less complex compared to existing high-level feature-based studies. The research outcomes demonstrate a promising pathway toward developing more robust, efficient, and scalable phishing detection systems. [ABSTRACT FROM AUTHOR]

Predicting the amount of soiling accumulated on the collectors is a key factor when optimizing the trade‐off between reducing soiling losses and cleaning costs. An important influence on soiling losses is natural cleaning through rain. Several soiling models assume complete cleaning through rain for daily rain sums above a model specific threshold and no cleaning otherwise. However, various studies show that cleaning is often incomplete. This study employs two statistical learning methods to model the cleaning effect of rain, aiming to achieve more accurate results than a simple totally cleaned/no cleaning answer while also considering other parameters besides the rain sum. The models are tested using meteorological and soiling data from 33 measurement stations in West Africa. Linear regression seems to be a good alternative for predicting the reduction in soiling levels after a rainfall. [ABSTRACT FROM AUTHOR]

Spatio-temporal characterization analysis plays a key role in spatio-temporal data mining tasks such as social relationship inference, traffic flow prediction, and spatio-temporal graph node classification. Although traditional numerical simulation methods are effective, they often struggle to accurately portray the complex characteristics of spatio-temporal data due to the intricacy of the modeling processes and the limitations of underlying assumptions. Models based on statistical learning and machine learning, such as ARIMA (AutoRegressive Integrated Moving Average Model) and SVM (Support Vector Machine), are capable of handling spatio-temporal data to a certain extent, but they are limited in their ability to characterize highly nonlinear data and can fail to effectively capture spatio-temporal correlations. To address these challenges, this paper introduces a new deep learning model, the Attention-Guided Spatio-Temporal Convolutional Network (AG-STCN). In the spatial dimension, the model captures spatial dependencies through an attention-guided soft pruning strategy and graph convolution operations. In the temporal dimension, it employs causal convolutions, gated linear units, and a self-attention mechanism to capture temporal dependencies. Experimental results demonstrate that the AG-STCN significantly outperforms existing baseline methods on real-world datasets for multiple spatio-temporal characterization analysis tasks. [ABSTRACT FROM AUTHOR]

Background: Snakebite envenoming is a serious condition that affects 2.5 million people and causes 81,000–138,000 deaths every year, particularly in tropical and subtropical regions. The World Health Organization has set a goal to halve the deaths and disabilities related to snakebite envenoming by 2030. However, significant challenges in achieving this goal include a lack of robust research evidence related to snakebite incidence and treatment, particularly in sub-Saharan Africa. This study aimed to combine established methodologies with the latest tools in Artificial Intelligence to assess the barriers to effective snakebite treatment in Ghana. Method: We used a MaxDiff statistical experiment design to collect data, and six supervised machine learning models were applied to predict responses whose performance showed an advantage over the other through 6921 data points partitioned using the hold-back validation method, with 70% training and 30% validation. The results were compared using key metrics: Akaike Information Criterion corrected, Bayesian Information Criterion, Root Average Squared Error, and Fit Time in milliseconds. Results: Considering all the responses, none of the six machine learning algorithms proved superior, but the Generalized Regression Model (Ridge) performed consistently better among the candidate models. The model consistently predicted several key significant barriers to effective snakebite treatment, such as the high cost of antivenoms, increased use of unorthodox, harmful practices, lack of access to effective antivenoms in remote areas when needed, and resorting to unorthodox and harmful practices in addition to hospital treatment. Conclusion: The combination of a MaxDiff statistical experiment design to collect data and six machine learning models allowed the identification of barriers to accessing effective therapies for snakebite envenoming in Ghana. Addressing these barriers through targeted policy interventions, including intensified advocacy, continuous education, community engagement, healthcare worker training, and strategic investments, can enhance the effectiveness of snakebite treatment, ultimately benefiting snakebite victims and reducing the burden of snakebite envenoming. There is a need for robust regulatory frameworks and increased antivenom production to address these barriers. Author summary: Snakebite envenoming, a severe condition, affects 2.5 million people and causes 81,000–138,000 deaths annually, particularly in tropical and subtropical regions of Africa, Asia, and Latin America. The World Health Organization aims to reduce this burden by 50% by 2030. However, significant barriers exist to achieving these targets, especially in Sub-Saharan Africa. These barriers include limited rigorous research evidence and the lack of investment in effective antivenoms for local snake species. Meeting this goal will require innovative research to gather better data on snakebite incidence and treatment. Artificial intelligence is one promising field that can contribute to this effort. For the first time, we have demonstrated how MaxDiff statistical experiment designs and machine learning algorithms can be explored to predict the barriers to effective snakebite treatment, such as the high cost of antivenoms, increased use of harmful practices, lack of access to effective antivenoms in remote areas, and resorting to unorthodox and harmful practices in addition to hospital treatment. Addressing these barriers through targeted policy interventions, including intensified advocacy, continuous education, community engagement, healthcare worker training, and strategic investments, can enhance the effectiveness of snakebite treatment. Robust regulatory frameworks and increased local antivenom production are also needed to address these barriers. [ABSTRACT FROM AUTHOR]

In this study, a comprehensive methodology combining machine learning and statistical analysis was employed to investigate alterations in the metabolite profiles, including lipids, of breast cancer tissues and their subtypes. By integrating biological and machine learning feature selection techniques, along with univariate and multivariate analyses, a notable lipid signature was identified in breast cancer tissues. The results revealed elevated levels of saturated and monounsaturated phospholipids in breast cancer tissues, consistent with external validation findings. Additionally, lipidomics analysis in both the original and validation datasets indicated lower levels of most triacylglycerols compared to non-cancerous tissues, suggesting potential alterations in lipid storage and metabolism within cancer cells. Analysis of cancer subtypes revealed that levels of PC 30:0 were relatively reduced in HER2(−) samples that were ER(+) and PR(+) compared to those that were ER(−) and PR(−). Conversely, HER2(+) tumors, which were ER(−) and PR(−), exhibited increased concentrations of PC 30:0. This increase could potentially be linked to the role of Stearoyl-CoA-Desaturase 1 in breast cancer. Comprehensive metabolomic analyses of breast cancer can offer crucial insights into cancer development, aiding in early detection and treatment evaluation of this devastating disease. [ABSTRACT FROM AUTHOR]

Learning to suppress location(s) where a distractor frequently occurs can improve search efficiency, known as distractor-location probability-cueing. However, the impact of the volatility of distractor occurrence – how often distractor-present and – absent events switch – remains poorly understood. To investigate this, we contrasted two volatility regimens in an additional-singleton search paradigm: a low-volatility environment in which distractor-present trials tended to occur in streaks, and a high-volatility environment with more frequent alternations. The distractor appeared 13 times more often at a designated frequent location than any rare locations. We replicated the probability-cueing effect, which was consistent across both volatilty conditions. Interestingly, the target-location effect – slower responses to a target at the frequent distractor location – was robust in the high-volatility condition, but non-significant in the low-volatility condition. We propose a suppression-thresholding account: the activation threshold of the saliency-triggered suppression mechanism is dynamically adjusted based on the volatility and local frequency of distractor occurrence. [ABSTRACT FROM AUTHOR]

Over the past three decades, incidental bycatch has been the single most frequent verified cause of death of the endangered Saimaa ringed seal (Pusa hispida saimensis). Spatial and temporal fishing closures have been enforced to mitigate bycatch, which is mainly caused by the gillnets of recreational fishers. In this study, we employed an array of statistical machine learning methods to recognize patterns of death and to evaluate the impacts of annual fishing closures (15th April–30th June) on the recovery of the Saimaa ringed seal population during 1991–2021. We additionally used the potential biological removal (PBR) procedure to assess bycatch sustainability. The study shows that gillnet restriction areas are reflected in the timing of juvenile bycatch mortality of the Saimaa ringed seal. In the 1990s, peak mortality occurred at the beginning of June, but as the restrictions expanded regionally in the 2000s, the peak shifted to the beginning of July. Longer temporal coverage of annual closures would have improved juvenile survival. The study also shows that estimated bycatch mortality is higher than observed: the estimated bycatch averaged approximately two unobserved bycatches per one observed bycatch. Despite the continuing bycatch mortality, a larger number of juveniles nowadays survive to the age of 15 months due to fishing closures, and the population (some 420 individuals) has increased an average 4% per year between 2017 and 2021. However, human-caused mortality limits (PBR) were exceeded by observed bycatch only, which could lead to population depletion in the long run. [ABSTRACT FROM AUTHOR]

The use of satellite altimetry data for monitoring the water level regime of rivers in Arctic regions is limited due to the negative effect of complex fluvial morphology and ice cover on altimetric radar measurements. The generation of time series of river water level consists of two main stages: (1) accurate geographic selection of satellite measurements over the river channel and (2) calculation of the average level for a given date after filtering outliers. This work is based on measurements from the European altimetry satellites Sentinel-3A and Sentinel-3B. The paper proposes a method for detection of aberrant values in altimetric measurements (outliers) acquired over a wide floodplain section of the Kolyma River. The method improved the accuracy of resulting satellite time series of water level by 0.04–1.59 m (or 4–85%) compared to the widely used standard statistical method of altimetric measurement filtering. The suggested method is based on the combination of three algorithms of different complexity: statistical (Mahalanobis distance), clustering (Density-Based Spatial Clustering of Applications with Noise (DBSCAN)), and machine learning (Isolating Forest) methods. In the combined approach, values classified as outliers by at least two algorithms were considered as outliers. This approach allowed us to reduce the impact of potential individual shortcomings of each of the three methods. [ABSTRACT FROM AUTHOR]

Our research explores the relationship between cognition and language. The focus of this paper is to discuss how we embarked upon remote data collection with children during the COVID-19 pandemic. In this study we investigate cognitive processes of non-verbal intelligence, working memory, implicit statistical learning, and speed of automatization (measured with the multiple-trial Tower of Hanoi puzzle). Here we focus primarily on the speed of automatization, partly because of theoretical interest, and because it is more difficult to adapt to an online format due to the motor component of the task. We established a hybrid method of data collection where the researcher was present online to guide children through a battery of language and cognitive tasks. We used a videoconferencing platform, a digital visualiser, and a physical puzzle which we posted to each child prior to commencing the research sessions. We also designed an online version of the puzzle with support from the Getting Data project. We discuss the methodology of our study and the lessons learned during remote data collection. [ABSTRACT FROM AUTHOR]

This paper presents a new approach in the field of probability-informed machine learning (ML). It implies improving the results of ML algorithms and neural networks (NNs) by using probability models as a source of additional features in situations where it is impossible to increase the training datasets for various reasons. We introduce connected mixture components as a source of additional information that can be extracted from a mathematical model. These components are formed using probability mixture models and a special algorithm for merging parameters in the sliding window mode. This approach has been proven effective when applied to real-world time series data for short- and medium-term forecasting. In all cases, the models informed by the connected mixture components showed better results than those that did not use them, although different informed models may be effective for various datasets. The fundamental novelty of the research lies both in a new mathematical approach to informing ML models and in the demonstrated increase in forecasting accuracy in various applications. For geophysical spatiotemporal data, the decrease in Root Mean Square Error (RMSE) was up to 27.7 % , and the reduction in Mean Absolute Percentage Error (MAPE) was up to 45.7 % compared with ML models without probability informing. The best metrics values were obtained by an informed ensemble architecture that fuses the results of a Long Short-Term Memory (LSTM) network and a transformer. The Mean Squared Error (MSE) for the electricity transformer oil temperature from the ETDataset had improved by up to 10.0 % compared with vanilla methods. The best MSE value was obtained by informed random forest. The introduced probability-informed approach allows us to outperform the results of both transformer NN architectures and classical statistical and machine learning methods. [ABSTRACT FROM AUTHOR]

Identifying characteristic patterns in time series, such as heartbeats or brain responses to a stimulus, is critical to understanding the physical or physiological phenomena monitored with sensors. Convolutional sparse coding (CSC) methods, which aim to approximate signals by a sparse combination of short signal templates (also called atoms), are well‐suited for this task. However, enforcing sparsity leads to non‐convex and untractable optimization problems. This article proposes finding the optimal solution to the original and non‐convex CSC problem when the atoms do not overlap. Specifically, we show that the reconstruction error satisfies a simple recursive relationship in this setting, which leads to an efficient detection algorithm. We prove that our method correctly estimates the number of patterns and their localization, up to a detection margin that depends on a certain measure of the signal‐to‐noise ratio. In a thorough empirical study, with simulated and real‐world physiological data sets, our method is shown to be more accurate than existing algorithms at detecting the patterns' onsets. [ABSTRACT FROM AUTHOR]

Deep learning has attracted widespread attention in data modeling and key quality indicator prediction in the chemical industry. However, traditional deep learning networks usually distort the original data distribution due to the superposition effect of multiple layers of nonlinear activation functions. In this case, multivariate statistical learning techniques present an avenue to reveal the intrinsic relationship of the data by combining the linear trends between input and predictor variables. To comprehensively capture data features from multiple perspectives, this study proposes a deep learning‐based data modeling network called the information retention unit (IRU). This network combines intrinsic attributes to partial least squares (PLS) and autoencoder (AE) modalities, thus engendering an adaptive response to the complex linear and nonlinear data features. Furthermore, multiple IRUs can be stacked to construct a deep information retention network (DIRN), which enhances the robust extraction of deep data features. Finally, the effectiveness of the proposed network is validated through its prediction application on a dataset obtained from a real‐world chemical industrial process. This method combines multivariate statistical learning techniques based on deep learning, providing an innovative and practical solution for data analysis and prediction in the chemical industry. [ABSTRACT FROM AUTHOR]

Summary: Model selection is a ubiquitous problem that arises in the application of many statistical and machine learning methods. In the likelihood and related settings, it is typical to use the method of information criteria (ICs) to choose the most parsimonious among competing models by penalizing the likelihood‐based objective function. Theorems guaranteeing the consistency of ICs can often be difficult to verify and are often specific and bespoke. We present a set of results that guarantee consistency for a class of ICs, which we call PanIC (from the Greek root 'pan', meaning 'of everything'), with easily verifiable regularity conditions. PanICs are applicable in any loss‐based learning problem and are not exclusive to likelihood problems. We illustrate the verification of regularity conditions for model selection problems regarding finite mixture models, least absolute deviation and support vector regression and principal component analysis, and demonstrate the effectiveness of PanICs for such problems via numerical simulations. Furthermore, we present new sufficient conditions for the consistency of BIC‐like estimators and provide comparisons of the BIC with PanIC. [ABSTRACT FROM AUTHOR]

Background: Dengue virus causes illnesses with or without warning indicators for severe complications. There are no clear prognostic signs linked to the disease outcomes. Methods: Clinical and laboratory parameters among 102 adult including 17 severe dengue (SD), 33 with warning and 52 without warning signs during early and critical phases were analysed by statistical and machine learning (ML) models. Results: In classical statistics, abnormal ultrasound findings, platelet count and low lymphocytes were significantly linked with SD during the febrile phase, while low creatinine, high sodium and elevated AST/ALT during the critical phase. ML models highlighted AST/ALT and lymphocytes as key markers for distinguishing SD from non-severe dengue, aiding clinical decisions. Conclusion: Parameters like liver enzymes, platelet counts and USG findings were linked with SD.USG testing at an earlier phase of dengue and a point-of-care system for the quantification of AST/ALT levels may lead to an early prediction of SD. [ABSTRACT FROM AUTHOR]

–Devising a dynamic pricing policy with always valid online statistical learning procedures is an important and as yet unresolved problem. Most existing dynamic pricing policies, which focus on the faithfulness of adopted customer choice models, exhibit a limited capability for adapting to the online uncertainty of learned statistical models during the pricing process. In this article, we propose a novel approach for designing a dynamic pricing policy based on regularized online statistical learning with theoretical guarantees. The new approach overcomes the challenge of continuous monitoring of the online Lasso procedure and possesses several appealing properties. In particular, we make the decisive observation that the always-validity of pricing decisions builds and thrives on the online regularization scheme. Our proposed online regularization scheme equips the proposed optimistic online regularized maximum likelihood pricing (OORMLP) pricing policy with three major advantages: encode market noise knowledge into pricing process optimism; empower online statistical learning with always-validity overall decision points; envelope prediction error process with time-uniform non-asymptotic oracle inequalities. This type of non-asymptotic inference results allows us to design more sample-efficient and robust dynamic pricing algorithms in practice. In theory, the proposed OORMLP algorithm exploits the sparsity structure of high-dimensional models and secures a logarithmic regret in a decision horizon. These theoretical advances are made possible by proposing an optimistic online Lasso procedure that resolves dynamic pricing problems at the process level, based on a novel use of non-asymptotic martingale concentration. In experiments, we evaluate OORMLP in different synthetic and real pricing problem settings and demonstrate that OORMLP advances the state-of-the-art methods. for this article are available online. [ABSTRACT FROM AUTHOR]

This article introduces a Factor Augmented Sparse Throughput (FAST) model that uses both latent factors and sparse idiosyncratic components for nonparametric regression. It contains many popular statistical models. The FAST model bridges factor models on one end and sparse nonparametric models on the other end. It encompasses structured nonparametric models such as factor augmented additive models and sparse low-dimensional nonparametric interaction models and covers the cases where the covariates do not admit factor structures. This model allows us to conduct high-dimensional nonparametric model selection for both strong dependent and weak dependent covariates and hence contributes to interpretable machine learning, particularly to the feature selections for neural networks. Via diversified projections as estimation of latent factor space, we employ truncated deep ReLU networks to nonparametric factor regression without regularization and to a more general FAST model using nonconvex regularization, resulting in factor augmented regression using neural network (FAR-NN) and FAST-NN estimators, respectively. We show that FAR-NN and FAST-NN estimators adapt to the unknown low-dimensional structure using hierarchical composition models in nonasymptotic minimax rates. We also study statistical learning for the factor augmented sparse additive model using a more specific neural network architecture. Our results are applicable to the weak dependent cases without factor structures. In proving the main technical result for FAST-NN, we establish a new deep ReLU network approximation result that contributes to the foundation of neural network theory. Numerical studies further support our theory and methods. for this article are available online. [ABSTRACT FROM AUTHOR]

Dynamic treatment regimen (DTR) is one of the most important tools to tailor treatment in personalized medicine. For many diseases such as cancer and type 2 diabetes mellitus (T2D), more aggressive treatments can lead to a higher efficacy but may also increase risk. However, few methods for estimating DTRs can take into account both cumulative benefit and risk. In this work, we propose a general statistical learning framework to learn optimal DTRs that maximize the reward outcome while controlling the cumulative adverse risk to be below a pre-specified threshold. We convert this constrained optimization problem into an unconstrained optimization using a Lagrange function. We then solve the latter using either backward learning algorithms or simultaneously over all stages based on constructing a novel multistage ramp loss. Theoretically, we establish Fisher consistency of the proposed method and further obtain non-asymptotic convergence rates for both reward and risk outcomes under the estimated DTRs. The finite sample performance of the proposed method is demonstrated via simulation studies and through an application to a two-stage clinical trial for T2D patients. for this article are available online. [ABSTRACT FROM AUTHOR]

Being able to process multiword sequences is central for both language comprehension and production. Numerous studies support this claim, but less is known about the way multiword sequences are acquired, and more specifically how associations between their constituents are established over time. Here we adapted the Hebb naming task into a Hebb lexical decision task to study the dynamics of multiword sequence extraction. Participants had to read letter strings presented on a computer screen and were required to classify them as words or pseudowords. Unknown to the participants, a triplet of words or pseudowords systematically appeared in the same order and random words or pseudowords were inserted between two repetitions of the triplet. We found that response times (RTs) for the unpredictable first position in the triplet decreased over repetitions (i.e., indicating the presence of a repetition effect) but more slowly and with a different dynamic compared with items appearing at the predictable second and third positions in the repeated triplet (i.e., showing a slightly different predictability effect). Implicit and explicit learning also varied as a function of the nature of the triplet (i.e., unrelated words, pseudowords, semantically related words, or idioms). Overall, these results provide new empirical evidence about the dynamics of multiword sequence extraction, and more generally about the role of statistical learning in language acquisition. [ABSTRACT FROM AUTHOR]

Crop yield prediction gains growing importance for all stakeholders in agriculture. Since the growth and development of crops are fully connected with many weather factors, it is inevitable to incorporate meteorological information into yield prediction mechanism. The changes in climate-yield relationship are more pronounced at a local level than across relatively large regions. Hence, district or sub-region-level modeling may be an appropriate approach. To obtain a location- and crop-specific model, different models with different functional forms have to be explored. This systematic review aims to discuss research papers related to statistical and machine-learning models commonly used to predict crop yield using weather factors. It was found that Artificial Neural Network (ANN) and Multiple Linear Regression were the most applied models. Support Vector Regression (SVR) model has a high success ratio as it performed well in most of the cases. The optimization options in ANN and SVR models allow us to tune models to specific patterns of association between weather conditions of a location and crop yield. ANN model can be trained using different activation functions with optimized learning rate and number of hidden layer neurons. Similarly, the SVR model can be trained with different kernel functions and various combinations of hyperparameters. Penalized regression models namely, LASSO and Elastic Net are better alternatives to simple linear regression. The nonlinear machine learning models namely, SVR and ANN were found to perform better in most of the cases which indicates there exists a nonlinear complex association between crop yield and weather factors. [ABSTRACT FROM AUTHOR]

PURPOSE: Patients with epithelial ovarian cancer (EOC) have an elevated risk for venous thromboembolism (VTE). To assess the risk of VTE, models were developed by statistical or machine learning algorithms. However, few models have accommodated deep learning (DL) algorithms in realistic clinical settings. We aimed to develop a predictive DL model, exploiting rich information from electronic health records (EHRs), including dynamic clinical features and the presence of competing risks. METHODS: We extracted EHRs of 1,268 patients diagnosed with EOC from January 2007 through December 2017 at the National Cancer Center, Korea. DL survival networks using fully connected layers, temporal attention, and recurrent neural networks were adopted and compared with multi-perceptron–based classification models. Prediction accuracy was independently validated in the data set of 423 patients newly diagnosed with EOC from January 2018 to December 2019. Personalized risk plots displaying the individual interval risk were developed. RESULTS: DL-based survival networks achieved a superior area under the receiver operating characteristic curve (AUROC) between 0.95 and 0.98 while the AUROC of classification models was between 0.85 and 0.90. As clinical information benefits the prediction accuracy, the proposed dynamic survival network outperformed other survival networks for the test and validation data set with the highest time-dependent concordance index (0.974, 0.975) and lowest Brier score (0.051, 0.049) at 6 months after a cancer diagnosis. Our visualization showed that the interval risk fluctuating along with the changes in longitudinal clinical features. CONCLUSION: Adaption of dynamic patient clinical features and accounting for competing risks from EHRs into the DL algorithms demonstrated VTE risk prediction with high accuracy. Our results show that this novel dynamic survival network can provide personalized risk prediction with the potential to assist risk-based clinical intervention to prevent VTE among patients with EOC. [ABSTRACT FROM AUTHOR]

How does media coverage of electoral campaigns distinguish parties and candidates in emerging democracies? To answer, we present a multi-step procedure that we apply in South Africa. First, we develop a theoretically informed classification of election coverage as either "narrow" or "broad" from within the entire corpus of news coverage during an electoral campaign. Second, to deploy our classification scheme, we use a supervised machine learning approach to classify news as "broad," "narrow," or "not election-related." Finally, we combine our supervised classification with a topic modeling algorithm (BERTTopic) that is based on Bidirectional Encoder Representations from Transformers (BERT), in addition to other statistical and machine learning methods. The combination of our classification scheme, BERTTopic, and associated methods allows us to identify the main election-related themes among broad and narrow election-related coverage, and how different candidates and parties are associated with these themes. We provide an in-depth discussion of our method for interested users in the social sciences. We then apply our proposed techniques on text from nearly 100,000 news articles during South Africa's 2014 campaign and test our empirical predictions about candidate and party coverage of corruption, the economy, health, public infrastructure, and security. The application of our method highlights a nuanced campaign environment in South Africa; candidates and parties frequently receive distinct and substantive coverage on key campaign themes. [ABSTRACT FROM AUTHOR]