Search

Traditional machine learning algorithms face challenges in assessing flood susceptibility reliably due to their low robustness and the inherent 'black-box' nature. This paper utilizes five hyperparameter optimization algoirthms (HPO), namely grid search (GS), random search (RS), gauss process (GP), tree-structured parzen estimator (TPE) and simulated annealing (SA), to tune the traditional random forest's (RF) hyperparameters to improve the robustness of flood hazard mapping (FHM) models at Ningxiang City Hunan Province, China. Additionally, SHapley Additive exPlanations (SHAP) method were used to interpret the decision-mechanisms of these flood hazard models. This study considers 19 pluvial flood influencing factors and 2064 flood locations to create a geospatial database. The performance of each hybrid model was evaluated by area under the receiver operating characteristic (ROC) curve (AUC) and several validation methods. The results demonstrate that the developed hybrid models demonstrated good performance, with RF-TPE achieving the highest AUC (0.9660), followed by RF-GP (0.9648), RF-SA (0.9624), RF-GS (0.9612), RF-RS (0.9600), and RF (0.9539). The RF-TPE model exhibits superior robustness than other models, and the FHM constructed using it is more reliable. HPO is an effective approach to improve the predictive accuracy and robustness of FHM models. When considering limited computational resources, Bayesian optimization (TPE) should be prioritized for optimizing FHM models, followed by metaheuristic algorithms and model-free algorithms. Moreover, the study revealed that distance from river, peak rainfall intensity, continuous rainfall, antecedent effective rainfall, and terrain relief, are the most significant for pluvial FHM modeling in this region. [ABSTRACT FROM AUTHOR]

Decreased vision acuity and blindness due to retinal detachment (RD) are significant concerns, emphasizing the importance of early diagnosis and identification. However, manual screening of RD is labor-intensive and time-consuming and faces challenges such as poor quality and low accuracy. A novel hybrid machine learning(ML) algorithm incorporating Levy flight-based atom search optimization (LFB-ASO) is proposed to solve the above challenges. The dataset utilized for the experiment is the Retinal Fundus Multi-disease Image dataset (RFMiD). The data preprocessing pipeline involves image resizing, normalization, data augmentation, masking and segmentation. To ensure consistent dimensions, all retinal images are standardized through resizing. Performance and convergence are improved using normalization. The data augmentation technique enhances diversity, while segmentation focuses on the region of interest (ROI). Then the deep features are extracted from the preprocessed retinal images using a pre-trained ResNet18 model. LFB-ASO is employed to select the most discriminative deep features for RD classification. To achieve superior accuracy, hybrid ML algorithms, namely Support Vector Machine (SVM), Gradient Boosting Machine (GBM) and Random Forest (RF) are employed. The proposed model achieves remarkable results with accuracy, recall, F1 score and precision of 98.75%, 96.70%, 97.01% and 97.62%. These results outperform existing methods such as HOS-LSDA, LR, NB, PCA and RD-Light-Net. [ABSTRACT FROM AUTHOR]

People with Parkinson's Disease (PwPD) usually experience several neuropsychiatric signs such as anxiety, depression, and negative emotions that contribute to disability and worsening of quality of life. Notwithstanding, the assessment of these symptoms are largely underrated, subjective and difficult due to a large overlapping with other PD symptoms, like hypomimia and bradikinesia. The aim and novelty of the current work is to study and validate a method for automatic emotion recognition in PwPD during daily living through autonomic signals acquired by acceptable and low-cost consumer technology. The best shallow learning algorithm and the best minimal feature set are individuated. 11 PwPD and 8 subjects with no history of neurological injury or illness were enrolled in the study. Participants were asked to watch video clips purposely selected to arouse emotions, and annotate arousal and valence of emotions triggered by video clips, while their heart rate, skin conductance, and temperature were recorded by a smartwatch. Smartwatch data was used for features extraction, while participants' reported arousal and valence were used as gold-standard to train machine learning algorithms for emotion classification (low/high arousal, positive/negative valence). Different feature sets and different algorithms (i.e. decision tree (DT), random forest (RF), support vector machine (SVM), and multilayer perceptron (MLP)) were evaluated to find the best solution for each group of participant. In each group of participants, it was possible to find a combination of feature set and algorithm to reach a classification accuracy greater than 90%. The Random Forest reached the best performance in both groups and for both valence and arousal. For each classification task (valence or arousal, PwPD or controls), the best model was selected and the minimal feature set was found by performing a recursive feature elimination based on the Shapley value. A lower accuracy of appraisal emerged for arousal compared to valence. Obtained results showed the feasibility of automatic emotion recognition in PwPDs through autonomic signals. Autonomic dysfunction in PwPDs may explain the lower arousal accuracy. The findings warrant confirmation from trials on larger samples and there are open issues to be deepened in future work. • Smartwatch based automatic emotion recognition (AER). • AER in Parkinson's disease through autonomic nervous system signals. • Comparison of shallow learning algorithms for AER in Parkinson's disease. • Analysis of emotional baseline and features normalization on performance. • SHAP-based recursive feature elimination on the best algorithm. [ABSTRACT FROM AUTHOR]

• We predict the number of shares of fake news shares based on its content. • The propagation of fake and real news follows different patterns. • Fake news that conveys strongly negative emotions are more likely to be shared. • We use machine learning to detect early fake news that are potentially viral. Fake news about brands and companies can cause reputation and financial damage, but there exist preventive strategies that could be implemented if we were able to early predict the virality of fake news pieces. In this paper, we consider the early prediction of fake news virality using textual information (body and headlines). We show that the impact of content features on virality is different for true and fake news, and so the propagation of true and fake news is. Furthermore, we use machine learning nonparametric models to classify fake news according to its propagation level. We also discuss defensive response strategies. [ABSTRACT FROM AUTHOR]

The aim of ordinal classification is to predict the ordered labels of the output from a set of observed inputs. Interval-valued data refers to data in the form of intervals. For the first time, interval-valued data and interval-valued functional data are considered as inputs in an ordinal classification problem. Six ordinal classifiers for interval data and interval-valued functional data are proposed. Three of them are parametric, one of them is based on ordinal binary decompositions and the other two are based on ordered logistic regression. The other three methods are based on the use of distances between interval data and kernels on interval data. One of the methods uses the weighted k -nearest-neighbor technique for ordinal classification. Another method considers kernel principal component analysis plus an ordinal classifier. And the sixth method, which is the method that performs best, uses a kernel-induced ordinal random forest. They are compared with naïve approaches in an extensive experimental study with synthetic and original real data sets, about human global development, and weather data. The results show that considering ordering and interval-valued information improves the accuracy. The source code and data sets are available at https://github.com/aleixalcacer/OCFIVD. • Six ordinal classifiers for interval data-valued data are proposed. • Ordinal methods for interval-valued functional data are also proposed. • Considering ordering and interval-valued information improves the accuracy. • A kernel-induced ordinal random forest performs best. [ABSTRACT FROM AUTHOR]

The cyber risk from malicious external attackers is a significant socio-economic problem. Cyber risk prediction is particularly difficult, given the constantly changing attack vectors. This study presents a model that automatically predicts cyber risks. The model is only based on common vulnerabilities and exposures (CVE) data and supervised prediction algorithms. This approach eliminates expert opinion bias in cyber risk prediction. Our supervised data-driven model, C y R i P r e d , CVE data into cyber risk groups by mapping the textual description field of the database into relevant Wikipedia article titles. Then C y R i P r e d aggregates the occurrence and severity of extracted topics for the desired time unit and produces a time series fed to supervised regressors for prediction. The risks are calculated using predicted occurrence and impact. Finally, the cyber risks are ranked by their score, and the top ten risks are presented. The proposed model is evaluated, and the results are discussed. • The model predicts cyber risks using CVE data, removing expert bias. • C y R i P r e d associates risks with relevant cybersecurity topics in real-time. • Novel W i k i T o p i c sub-module enhances risk categorisation with Wikipedia. [ABSTRACT FROM AUTHOR]

• Propose an improved random forest based on the improvement of decision trees. • Improve the evaluation mechanism for the classification effect of decision trees. • Propose a method for quantifying the diversity between decision trees. • Multiple tests verify the superiority of the proposed improved random forest. Random forest is one of the most widely used machine learning algorithms. Decision trees used to construct the random forest may have low classification accuracies or high correlations, which affects the comprehensive performance of the random forest. Aiming at these problems, the authors proposed an improved random forest based on the classification accuracy and correlation measurement of decision trees in this paper. Its main idea includes two parts, one is retaining the classification and regression trees (CARTs) with better classification effects, the other is reducing the correlations between the CARTs. Specifically, in the classification effect evaluation part, each CART was applied to make predictions on three reserved data sets, then the average classification accuracies were achieved, respectively. Thus, all the CARTs were sorted in descending order according to their achieved average classification accuracies. In the correlation measurement part, the improved dot product method was proposed to calculate the cosine similarity, i.e., the correlation, between CARTs in the feature space. By using the achieved average classification accuracy as reference, the grid search method was used to find the inner product threshold. On this basis, the CARTs with low average classification accuracy among CART pairs whose inner product values are higher than the inner product threshold were marked as deletable. The achieved average classification accuracies and correlations of CARTs were comprehensively considered, those with high correlation and weak classification effect were deleted, and those with better quality were retained to construct the random forest. Multiple experiments show that, the proposed improved random forest achieved higher average classification accuracy than the five random forests used for comparison, and the lead was stable. The G-means and out-of-bag data (OBD) score obtained by the proposed improved random forest were also higher than the five random forests, and the lead was more obvious. In addition, the test results of three non-parametric tests show that, there were significant diversities between the proposed improved random forest and the other five random forests. This effectively proves the superiority and practicability of the proposed improved random forest. [ABSTRACT FROM AUTHOR]

Gold has practical applications in jewellery, technology, investors and by central banks. This diversity of gold demand and self-balancing nature of the gold market underpin gold's robust qualities as an investment asset. Accurately forecasting gold price is conductive to protect and promote portfolio's performing. The extant studies, however, fail to capture the volatility in gold price from the perspective of analyzing and forecasting. In this study, a novel separation-ensemble method, which incorporates decomposition, separation, prediction and integration (DSPI) approaches, is proposed to analyze and predict the international gold price. In particular, the proposed DSPI method begins with extreme-point symmetric mode decomposition (ESMD) for retrieving the intrinsic mode functions (IMFs), which is used to generate the multi-channel mixed signals of the gold price. Next, hierarchical agglomerative clustering (HAC) algorithm is utilized to reconstruct the IMFs. Additionally, the recombination of IMFs is regarded as input of separation module and independent components (ICs) can be separated by the RLS-type independent component analysis (RLS-ICA). Especially, we compare the interactive relationship between the gold price and underlying meanings of the ICs. Finally, the proposed heterogeneous forecasting scheme based on induced ordered weighted averaging (IOWA) approach, in the sense that it assigns a diverse contribution to the forecasting results of the ICs predicted by the individual forecasting approaches, is utilized to achieve the forecasting result of the gold price. Experimental results on the gold price collected from World Gold Council confirm the effectiveness of the proposed DSPI approach, and the comparison results show that it is superior to the extant combination methods. • A novel separation-ensemble approach is developed based on RLS-type ICA. • The gold price is analyzed in terms of underlying factors. • Heterogeneous prediction scheme is proposed for forecasting gold price. • The superiority of the proposed DSPI model is verified via performance criteria. [ABSTRACT FROM AUTHOR]

The detection of epileptic seizures in EEG signals is a challenging task because it requires careful review of multi-channel EEG recordings over a lengthy time interval. In general, EEG-based seizure detection is strongly dependent on the ability to select descriptive features that are also stable in the sense that they are not sensitive to changes in the training data. This study proposes and investigates a patient-independent seizure detection model that uses stable EEG-based features obtained by comparing multiple feature selection methods. The schemes considered can be divided into five categories, often referred to as similarity, information theoretic, sparse learning, statistical, and graph centrality feature selection methods. The stability of the features these schemes produce was evaluated by random forest classification, using the intersection, frequency, correlation, and similarity measures. In experiments with Temple University Hospital's Seizure database of EEG records of 341 patients, unsupervised graph centrality feature selection was the most effective method, with a correct classification rate of 91.5%. • This study introduces a patient-independent method for detecting epileptic seizures. • Stability of features selection methods is considered. • Highest stability obtained via an unsupervised graph-based feature selection method. • The stable selected features were mostly extracted from the F8 and T3 channels. • The proposed method achieves 91.5% classification accuracy. [ABSTRACT FROM AUTHOR]

In machine learning, some models can make uncertain and imprecise predictions, they are called evidential models. These models may also be able to handle imperfect labeling and take into account labels that are richer than the commonly used hard labels, containing uncertainty and imprecision. This paper proposes an Evidential Decision Tree, and an Evidential Random Forest. These two models use a distance and a degree of inclusion to allow the model to group observations whose response elements are included in each other into a single node. Experimental results showed better performance for the presented methods compared to other evidential models and to recent Cautious Random Forests when the data is noisy. The models also offer a better robustness to the overfitting effect when using datasets that are effectively uncertainly and imprecisely labeled by the contributors. The proposed models are also able to predict rich labels, an information that can be used in other approaches, such as active learning. • Source uncertainty and imprecision can be modeled in richer labels. • Richer labels improve classification performance. • Evidential decision tree and evidential random forest are robust to overfitting. [ABSTRACT FROM AUTHOR]

Accurate and reliable runoff prediction is of great significance to water resources management, disaster monitoring and rational development and utilization of water resources. In this paper, a metaheuristic evolutionary deep learning model based on Temporal Convolutional Network (TCN), Improved Aquila Optimizer (IAO) and Random Forest (RF) is proposed for rainfall-runoff simulation and multi-step runoff prediction. In this study, the influence of various input variables on the prediction accuracy is discussed. First of all, in order to avoid the dimensional disaster problems and reduce the calculation time, RF is used to calculate the correlation between the input variables and the prediction object, and the data with high correlation is selected as the final inputs. Then, the filtered data are sent to the TCN model, and the parameters of the TCN model are optimized using the IAO algorithm, and the final prediction results are obtained. In this study, the rainfall and runoff data of five stations in the middle reaches of Jinsha River, China were selected, and the runoff of Panzhihua station was simulated and predicted by establishing multiple models. By analyzing and comparing the predictive results of several models, it shows that the models and improvements proposed in this study are effective. [ABSTRACT FROM AUTHOR]

Accurate and reliable runoff prediction is of great significance to water resources management, disaster monitoring and rational development and utilization of water resources. In this paper, a metaheuristic evolutionary deep learning model based on Temporal Convolutional Network (TCN), Improved Aquila Optimizer (IAO) and Random Forest (RF) is proposed for rainfall-runoff simulation and multi-step runoff prediction. In this study, the influence of various input variables on the prediction accuracy is discussed. First of all, in order to avoid the dimensional disaster problems and reduce the calculation time, RF is used to calculate the correlation between the input variables and the prediction object, and the data with high correlation is selected as the final inputs. Then, the filtered data are sent to the TCN model, and the parameters of the TCN model are optimized using the IAO algorithm, and the final prediction results are obtained. In this study, the rainfall and runoff data of five stations in the middle reaches of Jinsha River, China were selected, and the runoff of Panzhihua station was simulated and predicted by establishing multiple models. By analyzing and comparing the predictive results of several models, it shows that the models and improvements proposed in this study are effective. [ABSTRACT FROM AUTHOR]

Flight delay prediction has become one of the most critical topics in intelligent aviation systems due to its essential role in flight scheduling, airline planning, and airport operation. The accurate prediction of flight delays is very challenging because numerical factors will affect flight delays. Moreover, owing to the connectivity of the aviation system, flight delays also present complex spatial–temporal correlations, including the spatial correlations between airports along with the temporal correlations among timestamps. To address these challenges, we proposed a CNN-LSTM deep learning framework to consider the spatial–temporal correlations together with the extrinsic features for flight delay prediction. The CNN-LSTM model consists of a Convolution neural network (CNN) architecture to learn the spatial correlations followed by a Long short-term memory (LSTM) architecture to capture the temporal correlations. The spatial–temporal correlations obtained from the CNN-LSTM framework are then fused with the extrinsic features (e.g., airline issues, distance, schedule fly time, etc.) as inputs of the random forest (RF) model for flight delay prediction. The U.S. domestic flights in 2019 are collected from the Bureau of Transport Statistics to confirm the outperformance of the proposed model. The results show that the accuracy of the CNN-LSTM model reaches 92.39%. For the on-time samples, approximately 91% are correctly identified; for the delayed samples, the classification accuracy reaches 84% which exhibits better performance compared with several benchmark models. The created prediction model of this study could provide useful information for airport regulators in understanding the potential delays in advance and developing effective airport management strategies for improving airport on-time performance. [ABSTRACT FROM AUTHOR]

• Black-box decision-making algorithms proposals can be counter-intuitive. • Existing interpretability tools only extract global statistics. • Focus on interpretability for singular cases needed for data-driven decision making. • The algorithm is easy to implement, efficient and model-agnostic. With the widespread adoption of data mining models to solve real-world problems, the scientific community is facing the need of increasing their interpretability and comprehensibility. This is especially relevant in the case of black box models, in which inputs and outputs are usually connected by highly complex and nonlinear functions; in applications requiring an interaction between the user and the model; and when the machine's solution disagrees with the human experience. In this contribution we present a new methodology that allows to simplify the process of understanding the rules behind a classification model, even in the case of black box ones. It is based on the perturbation of the features describing one instance, and on finding the minimal variation required to change the forecasted class. It thus yields simplified rules describing under which circumstances would the solution have been different, and allows to compare these with the human expectation. We show how such methodology is well defined, model-agnostic, easy to implement and modular; and demonstrate its usefulness with several synthetic and real-world data sets. [ABSTRACT FROM AUTHOR]

• A distributed predictive model is proposed for Big Data recommendations. • A data partitioning strategy is adopted to speed up Big Data processing. • A novel learning process is devised to accelerate parallel and distributed training. • Matrix factorization and Random forest are used to improve the prediction accuracy. • Our proposal is highly competitive compared with state-of-the-art approaches. Recommendation systems have been widely deployed to address the challenge of overwhelming information. They are used to enable users to find interesting information from a large volume of data. However, in the era of Big Data, as data become larger and more complicated, a recommendation algorithm that runs in a traditional environment cannot be fast and effective. It requires a high computational cost for performing the training task, which may limit its applicability in real-world Big Data applications. In this paper, we propose a novel distributed recommendation solution for Big Data. It is designed based on Apache Spark to handle large-scale data, improve the prediction quality, and address the data sparsity problem. In particular, thanks to a novel learning process, the model is able to significantly speed up the distributed training, as well as improve the performance in the context of Big Data. Experimental results on three real-world data sets demonstrate that our proposal outperforms existing recommendation methods in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and computational time. [ABSTRACT FROM AUTHOR]

• We compare performance for random forest variable selection methods. • VSURF or Jiang's method are preferable for most datasets. • varSelRF or Boruta perform well for data with >50 predictors. • Methods with conditional random forest usually have similar performance. • Type of methods, test- or performance-based, is not likely to impact performance. Random forest classification is a popular machine learning method for developing prediction models in many research settings. Often in prediction modeling, a goal is to reduce the number of variables needed to obtain a prediction in order to reduce the burden of data collection and improve efficiency. Several variable selection methods exist for the setting of random forest classification; however, there is a paucity of literature to guide users as to which method may be preferable for different types of datasets. Using 311 classification datasets freely available online, we evaluate the prediction error rates, number of variables, computation times and area under the receiver operating curve for many random forest variable selection methods. We compare random forest variable selection methods for different types of datasets (datasets with binary outcomes, datasets with many predictors, and datasets with imbalanced outcomes) and for different types of methods (standard random forest versus conditional random forest methods and test based versus performance based methods). Based on our study, the best variable selection methods for most datasets are Jiang's method and the method implemented in the VSURF R package. For datasets with many predictors, the methods implemented in the R packages varSelRF and Boruta are preferable due to computational efficiency. A significant contribution of this study is the ability to assess different variable selection techniques in the setting of random forest classification in order to identify preferable methods based on applications in expert and intelligent systems. [ABSTRACT FROM AUTHOR]

• Credit products are the crucial part of business of retail banks. • Time series approach - events in database are processed as time series. • Machine learning system selects most promising customers for a marketing campaign. • Transactional data is used by the proposed system to extract significant patterns. Credit products are a crucial part of business of banks and other financial institutions. A novel approach based on time series of customer's data representation for predicting willingness to take a personal loan is shown. Proposed testing procedure based on moving window allows detection of complex, sequential, time based dependencies between particular transactions. Moreover, this approach reduces noise by eliminating irrelevant dependencies that would occur due to the lack of time dimension analysis. The system for identifying customers interested in credit products, based on classification with random forests and deep neural networks is proposed. The promising results of empirical studies prove that the system is able to extract significant patterns from customers historical transfer and transactional data and predict credit purchase likelihood. Our approach, including the testing method, is not limited to banking sector and can be easily transferred and implemented as a general purpose direct marketing campaign system. [ABSTRACT FROM AUTHOR]

Random Forest is an ensemble classification approach. It aims to design a discrete finite group of decision trees constructed based on bootstrap samples and random attribute selection. Random Forests have strong generalization capacities due to the variance in the training and attribute couple subsets used for constructing different decision trees in the forest. However, to construct a robust and effective random forest, two main issues need to be taken into account namely: (1) increasing the accuracy and diversity of decision trees; (2) decreasing the number of decision trees. In this paper, a genetic algorithm-based approach to tackle the aforementioned challenges related to random forest construction is proposed. Three objectives are taken into consideration. First, strengthening the classification accuracy of individual decision trees as well as that of the forest. Second, making use of diversity measures among the decision trees to improve the generalization of the constructed model. Third, minimizing the number of trees in the forest and finding an optimal subset of the random forest. An experimental evaluation on several datasets from the UCI Machine Learning Repository is conducted. The obtained results show that the proposed approach outperforms state-of-the-art classical as well as evolutionary random forest construction methods. Finally, the proposed approach is used to build a reliable random forest model for detecting Botnet traffic in Internet of Things environment. • We propose a genetic algorithm to construct an improved random forest. • We design a theoretic solution allowing a smart convergence to the best solutions. • We propose a Powerset system to optimize the encoding of the forests in chromosomes. • Our objective function optimizes the accuracy and diversity of the trees of models. • Application to the detection of Botnet traffic in Internet of Things networks. [ABSTRACT FROM AUTHOR]

The gradual depletion of surface water in major lakes and their impact in the sustainable development of local water resources has been a great challenge. Monitoring surface water and detecting changes in the lake are the main objectives of this study. Landsat Thematic Mapper (TM), Landsat Enhanced Thematic Mapper (ETM+) and Landsat Operational Land Imager (OLI) of 2010, 2000 and 2018 Lake Urmia images acquired from US Geological Survey were used for detecting the changes. Surface water changes are usually identified by extracting the water features from individual time series multispectral images. In this study, a novel change detection framework has been proposed involving pixel level fusion and classification. The spatial frequency based undecimated wavelet transform fusion (UDWT – SF) effectively extracted the spectral information from MS image and spatial information from PAN image of the same scene captured at different time periods. The endmembers of the fused images were selected using pixel purity index endmember extraction algorithm and the abundance estimation by the application of fully constrained least square spectral unmixing algorithm. An efficient sub-pixel classification process is designed by employing the spectral signatures, Normalized Difference Water Index (NDWI) and the abundance estimation generated from the pansharpened image in a random forest classifier. Experimental results indicate that the proposed classifier attained 99.9945% user's accuracy for water area and 99.9675% producer's accuracy for changed area. Similarly, the producer's accuracy of water and changed area are 99.9866% and 99.9868% respectively. The kappa coefficient and the overall accuracy of the proposed sub-pixel random forest classification on multitemporal multispectral image is 0.97 and 99.89% respectively. The lake surface area is computed and it is found that an area of 1369 Sq km has been decreased from the year 2000 to 2010 and 310 Sq km decreased from 2010 to 2018 as per the assessment based on the proposed random forest classifier. Spectral unmixing based random forest classifier (RF-SP) on fused image yields better results in terms of accuracy compared to other classifiers and it is more appropriate for detecting the multitemporal surface water changes. [ABSTRACT FROM AUTHOR]

Members of the armed forces rely on having an effective and efficient medical evacuation (MEDEVAC) process for evacuating casualties from the battlefield to medical treatment facilities during combat operations. Determining which MEDEVAC units to task to respond to incoming requests for service, known as the MEDEVAC dispatching problem, is a natural sequential decision making problem and is the focus of this paper. Unlike previous research in this area, this paper incorporates triage classification errors and the possibility of having blood transfusion kits on board select MEDEVAC units. A discounted, infinite-horizon continuous-time Markov decision process model is formulated to examine the MEDEVAC dispatching problem and to compare generated dispatching policies to the currently practiced United States military policy. We utilize an approximate dynamic programming (ADP) technique that leverages a random forest value function approximation within an approximate policy iteration algorithmic framework to develop high-quality policies for both small-scale and large-scale problem instances. A representative planning scenario involving joint combat operations in South Korea is developed and utilized to investigate the differences between the various policies. Results from the analysis indicate that applying ADP techniques can improve current practices by as much as 29% with regard to a life-saving performance metric. This research is of particular interest to the military medical community and can inform the procedures of future military MEDEVAC operations. [ABSTRACT FROM AUTHOR]

The concentrations of fine particulate matter (PM 2.5) constituents, which are very important and essential information for the identification of air pollution sources, were predicted at three sites (Seoul, Ulsan, Baengnyeong) in South Korea between 2016 and 2018 using four machine learning (ML) models: generative adversarial imputation network (GAIN), fully connected deep neural network (FCDNN), random forest (RF), and k-nearest neighbor (kNN). 3 PM 2.5 constituent groups, namely 8 ions, 2 carbons, and 15 trace elements, were targeted for prediction. The latest hyperparameter optimization techniques were used to learn air pollution characteristics from ambient PM 2.5 -related information, such as time, meteorology, and air pollutant concentrations. We compared the feature extraction abilities of the four models. The prediction accuracy identified by the coefficient of determination (R2) between prediction and observation was highest in GAIN, followed by FCDNN and RF or kNN. On availability of data on the time, air pollutant concentrations, and/or meteorology, simultaneously missed 20 % data of all PM 2.5 constituent groups were predicted, with R2 = 0.897, 0.861, 0.785, and 0.801 by the GAIN, FCDNN, RF, and kNN, respectively. As missing ratios (20 %, 40 %, 60 %, 80 %) of input data increased, prediction accuracy decreased in the four models and was predominantly more noticeable in GAIN and kNN. As the available period of data increased, prediction accuracy increased in GAIN and FCDNN. Trace elements were predicted with the lowest R2 in all models among the target constituent groups. Study sites with more emission sources showed lower prediction accuracy, resulting in the highest R2 in Baengnyeong island and the lowest in Ulsan. According to the current findings, ML models can be used to evaluate various air pollution issues for which data is missing. [ABSTRACT FROM AUTHOR]

Highlights • Detection of algorithmically generated domains using masked N-grams is introduced. • Dataset of algorithmically generated domains of real malware is publicly released. • Malware families are classified according to their domain generation algorithm. • Masked N-grams provide a good trade-off between training time and accuracy. Abstract Malware detection is a challenge that has increased in complexity in the last few years. A widely adopted strategy is to detect malware by means of analyzing network traffic, capturing the communications with their command and control (C&C) servers. However, some malware families have shifted to a stealthier communication strategy, since anti-malware companies maintain blacklists of known malicious locations. Instead of using static IP addresses or domain names, they algorithmically generate domain names that may host their C&C servers. Hence, blacklist approaches become ineffective since the number of domain names to block is large and varies from time to time. In this paper, we introduce a machine learning approach using Random Forest that relies on purely lexical features of the domain names to detect algorithmically generated domains. In particular, we propose using masked N-grams , together with other statistics obtained from the domain name. Furthermore, we provide a dataset built for experimentation that contains regular and algorithmically generated domain names, coming from different malware families. We also classify these families according to their type of domain generation algorithm. Our findings show that masked N-grams provide detection accuracy that is comparable to that of other existing techniques, but with much better performance. [ABSTRACT FROM AUTHOR]

Label ranking aims to learn a mapping from instances to rankings over a finite number of predefined labels. Random forest is a powerful and one of the most successful general-purpose machine learning algorithms of modern times. In this paper, we present a powerful random forest label ranking method which uses random decision trees to retrieve nearest neighbors. We have developed a novel two-step rank aggregation strategy to effectively aggregate neighboring rankings discovered by the random forest into a final predicted ranking. Compared with existing methods, the new random forest method has many advantages including its intrinsically scalable tree data structure, highly parallel-able computational architecture and much superior performance. We present extensive experimental results to demonstrate that our new method achieves the highly competitive performance compared with state-of-the-art methods for datasets with complete ranking and datasets with only partial ranking information. [ABSTRACT FROM AUTHOR]

Most machine learning algorithms possess hyperparameters. For example, an artificial neural network requires the determination of the number of hidden layers, nodes, and many other parameters related to the model fitting process. Despite this, there is still no clear consensus on how to tune them. The most popular methodology is an exhaustive grid search, which can be highly inefficient and sometimes infeasible. Another common solution is to change one hyperparameter at a time and measure its effect on the model’s performance. However, this can also be inefficient and does not guarantee optimal results since it ignores interactions between the hyperparameters. In this paper, we propose to use the Design of Experiments (DOE) methodology (factorial designs) for screening and Response Surface Methodology (RSM) to tune a machine learning algorithm’s hyperparameters. An application of our methodology is presented with a detailed discussion of the results of a random forest case-study using a publicly available dataset. Benefits include fewer training runs, better parameter selection, and a disciplined approach based on statistical theory. [ABSTRACT FROM AUTHOR]

Usage of depth data in various computer vision applications gains popularity with cheap depth sensors available on the market. In this study, a method that utilizes Deep Learning is proposed for single human actions and dyadic actions. Depth data of the actions are used to construct a three-dimensional (3D) template. These templates are rotated in different directions and two-dimensional (2D) views from different angles are stored. Acquired 2D images are used in deep feature extraction. AlexNet (Krizhevsky, Sutskever, & Hinton, 2012) pre-trained convolutional neural network is used for deep feature extraction. For all viewpoints, deep features are extracted and concatenated. After that random forest classifier is used for action recognition. The contributions of this paper are as follows. First, a 3D isosurface models is constructed to figure actions sequence. Second, we project 3D shapes into a 2D feature space by taking its snapshots from different views and giving them to a pre-trained CNN as input for feature extraction. We explore more information about the actions by extracting features from different layers of the deep neural network. We compare the results of the deep features extracted from different layers of the pre-trained CNN. Various classifiers are trained with extracted deep features. Because of the complex structure of the 3D shapes, there are a limited number of feature extraction methods. Our method obtains close to state-of-the-art performance on various single and two-person action datasets. [ABSTRACT FROM AUTHOR]

Epilepsy is the second common brain disorder affecting 70 million people worldwide. Electroencephalogram (EEG) has been widely used for the diagnosis of epileptic seizures. However, most of the existing EEG-based seizure detection methods cannot maintain robust performance in real life conditions. This is where the EEG data are corrupted with different sources of noise and artifacts. This study presents a robust seizure detection system that works efficiently under the real life conditions as well as the ideal conditions. A feature learning method based on L1-penalized robust regression is developed and applied to the EEG spectra to recognize the most prominent features pertinent to epileptic seizures. The extracted features are then fed into the random forest classifier for seizure detection. Results on a public benchmark dataset show that the performance of this seizure detection system is superior to prior work. It first achieves seizure detection rates of 100.00% sensitivity, 100.00% specificity, and 100.00% classification accuracy under the ideal conditions. The proposed method is also proven to be robust in the presence of white noise and EEG artifacts mainly those arising from muscle activities and eyes-blinking. It is found to achieve seizure detection accuracies in the range of 90.00–100.00% when applied to noisy EEG data corrupted with high noise levels. To the best of our knowledge, there exists no work in the literature that detects seizures under these conditions. [ABSTRACT FROM AUTHOR]

• Early identification of PCOS with the commonly known diseases. • Amalgamate the different diseases datasets for new dataset of PCOS. • Supervised learning algorithms are applied to the newly created dataset. • Unsupervised learning validates the outcomes of supervised learning algorithms. • Helpful for healthcare workers in the early detection of PCOS. Polycystic Ovary Syndrome (PCOS) is a health disorder that affects around 10 million women worldwide. Due to a lack of awareness in India, it affects one out of every-five women. If not diagnosed at an early stage, it leads to various other harmful diseases as well like diabetes, high blood pressure, obesity, heart disease etc. Thus, PCOS identification at an early stage is essential. Although several expensive tests are available there is no proper treatment and people are not aware of the same. The main objective of this article is to find the diseases which can help early identification of PCOS in a nutshell, we want to answer the question "Can we identify some commonly known diseases which taken as an indication of having PCOS". To achieve this objective, data amalgamation is performed on the identified disease datasets ahead and feature selection methods are applied resulting in 8 parameters and 985 records. To validate the desired hidden relation of PCOS with other diseases, supervised and unsupervised learning algorithms are applied. This article also tries to answer only important features of other diseases that are required for PCOS prediction. The performance metrics analysis with all features, important features and remaining features shows that the important features give the best result for PCOS identification. [ABSTRACT FROM AUTHOR]

This paper proposes a novel stock price trend prediction system that can predict both stock price movement and its interval of growth (or decline) rate within the predefined prediction durations. It utilizes an unsupervised heuristic algorithm to cut raw transaction data of each stock into multiple clips with the predefined fixed length and classifies them into four main classes ( Up, Down, Flat , and Unknown ) according to the shapes of their close prices. The clips in Up and Down can be further classified into different levels reflecting the extents of their growth (or decline) rates with respect to both close price and relative return rate. The features of clips include their prices and technical indices. The prediction models are trained from these clips by a combination of random forests, imbalance learning and feature selection. Evaluations on the seven-year Shenzhen Growth Enterprise Market (China) transaction data show that the proposed system can make effective predictions, is robust to the market volatility, and outperforms some existing methods in terms of accuracy and return per trade. [ABSTRACT FROM AUTHOR]

Random Forest (RF) learning algorithm is considered a classifier of reference due its excellent performance. Its success is based on the diversity of rules generated from decision trees that are built via a procedure that randomizes instances and features. To find additional procedures for increasing the diversity of the trees is an interesting task. It has been considered a new split criterion, based on imprecise probabilities and general uncertainty measures, that has a clear dependence of a parameter and has shown to be more successful than the classic ones. Using that criterion in RF scheme, join with a random procedure to select the value of that parameter, the diversity of the trees in the forest and the performance are increased. This fact gives rise to a new classification algorithm, called Random Credal Random Forest (RCRF). The new method represents several improvements with respect to the classic RF: the use of a more successful split criterion which is more robust to noise than the classic ones; and an increasing of the randomness which facilitates the diversity of the rules obtained. In an experimental study, it is shown that this new algorithm is a clear enhancement of RF, especially when it applied on data sets with class noise, where the standard RF has a notable deterioration. The problem of overfitting that appears when RF classifies data sets with class noise is solved with RCRF. This new algorithm can be considered as a powerful alternative to be used on data with or without class noise. [ABSTRACT FROM AUTHOR]

This work addresses classification of unknown binaries executed in sandbox by modeling their interaction with system resources (files, mutexes, registry keys and communication with servers over the network) and error messages provided by the operating system, using vocabulary-based method from the multiple instance learning paradigm. It introduces similarities suitable for individual resource types that combined with an approximative clustering method efficiently group the system resources and define features directly from data. This approach effectively removes randomization often employed by malware authors and projects samples into low-dimensional feature space suitable for common classifiers. An extensive comparison to the state of the art on a large corpus of binaries demonstrates that the proposed solution achieves superior results using only a fraction of training samples. Moreover, it makes use of a source of information different than most of the prior art, which increases the diversity of tools detecting the malware, hence making detection evasion more difficult. [ABSTRACT FROM AUTHOR]

In this paper, we propose a new decision forest algorithm that builds a set of highly accurate decision trees by exploiting the strength of all non-class attributes available in a data set, unlike some existing algorithms that use a subset of the non-class attributes. At the same time to promote strong diversity, the proposed algorithm imposes penalties (disadvantageous weights) to those attributes that participated in the latest tree in order to generate the subsequent trees. Besides, some other weight-related concerns are taken into account so that the trees generated by the proposed algorithm remain individually accurate and retain strong diversity. In order to show the worthiness of the proposed algorithm, we carry out experiments on 20 well known data sets that are publicly available from the UCI Machine Learning Repository. The experimental results indicate that the proposed algorithm is effective in generating highly accurate and more balanced decision forests compared to other prominent decision forest algorithms. Accordingly, the proposed algorithm is expected to be very effective in the domain of expert and intelligent systems. [ABSTRACT FROM AUTHOR]

Exploring relations between patterns of peak rotational speed of thorax, pelvis and arm, and patterns of EMG signals recorded from eight muscle regions of forearms and shoulders during the golf swing is the main objective of this article. The linear canonical correlation analysis, allowing studying relations between sets of variables, was the main technique applied. To get deeper insights, linear and nonlinear random forests-based prediction models relating a single output variable, e.g. a thorax peak rotational speed, with a set of input variables, e.g. an average intensity of EMG signals were used. The experimental investigations using data from 16 golfers revealed statistically significant relations between sets of input and output variables. A strong direct linear relation was observed between linear combinations of EMG averages and peak rotational speeds. The coefficient of determination values R 2 = 0.958 and R 2 = 0.943 obtained on unseen data by the random forest models designed to predict peak rotational speed of thorax and pelvis , indicate high modelling accuracy. However, predictions of peak rotational speed of arm were less accurate. This was expected, since peak rotational speed of arm played a minor role in the linear combination of peak speeds. The most important muscles to predict peak rotational speed of the body parts were identified. The investigations have shown that the canonical correlation analysis is a promising tool for studying relations between sets of biomechanical and EMG data. Better understanding of these relations will lead to guidelines concerning muscle engagement and coordination of thorax, pelvis and arms during a golf swing and will help golf coaches in providing substantiated advices. [ABSTRACT FROM AUTHOR]

Detecting seizure using brain neuroactivations recorded by intracranial electroencephalogram (iEEG) has been widely used for monitoring, diagnosing, and closed-loop therapy of epileptic patients, however, computational efficiency gains are needed if state-of-the-art methods are to be implemented in implanted devices. We present a novel method for automatic seizure detection based on iEEG data that outperforms current state-of-the-art seizure detection methods in terms of computational efficiency while maintaining the accuracy. The proposed algorithm incorporates an automatic channel selection (ACS) engine as a pre-processing stage to the seizure detection procedure. The ACS engine consists of supervised classifiers which aim to find iEEG channels which contribute the most to a seizure. Seizure detection stage involves feature extraction and classification. Feature extraction is performed in both frequency and time domains where spectral power and correlation between channel pairs are calculated. Random Forest is used in classification of interictal, ictal and early ictal periods of iEEG signals. Seizure detection in this paper is retrospective and patient-specific. iEEG data is accessed via Kaggle, provided by International Epilepsy Electro-physiology Portal. The dataset includes a training set of 6.5 h of interictal data and 41 min in ictal data and a test set of 9.14 h. Compared to the state-of-the-art on the same dataset, we achieve 2 times faster in run-time seizure detection. The proposed model is able to detect a seizure onset at 89.40% sensitivity and 89.24% specificity with a mean detection delay of 2.63 s for the test set. The area under the ROC curve ( AUC ) is 96.94%, that is comparable to the current state-of-the-art with AUC of 96.29%. [ABSTRACT FROM AUTHOR]

There has been intensive research from academics and practitioners regarding models for predicting bankruptcy and default events, for credit risk management. Seminal academic research has evaluated bankruptcy using traditional statistics techniques (e.g. discriminant analysis and logistic regression) and early artificial intelligence models (e.g. artificial neural networks). In this study, we test machine learning models (support vector machines, bagging, boosting, and random forest) to predict bankruptcy one year prior to the event, and compare their performance with results from discriminant analysis, logistic regression, and neural networks. We use data from 1985 to 2013 on North American firms, integrating information from the Salomon Center database and Compustat, analysing more than 10,000 firm-year observations. The key insight of the study is a substantial improvement in prediction accuracy using machine learning techniques especially when, in addition to the original Altman’s Z-score variables, we include six complementary financial indicators. Based on Carton and Hofer (2006), we use new variables, such as the operating margin, change in return-on-equity, change in price-to-book, and growth measures related to assets, sales, and number of employees, as predictive variables. Machine learning models show, on average, approximately 10% more accuracy in relation to traditional models. Comparing the best models, with all predictive variables, the machine learning technique related to random forest led to 87% accuracy, whereas logistic regression and linear discriminant analysis led to 69% and 50% accuracy, respectively, in the testing sample. We find that bagging, boosting, and random forest models outperform the others techniques, and that all prediction accuracy in the testing sample improves when the additional variables are included. Our research adds to the discussion of the continuing debate about superiority of computational methods over statistical techniques such as in Tsai, Hsu, and Yen (2014) and Yeh, Chi, and Lin (2014). In particular, for machine learning mechanisms, we do not find SVM to lead to higher accuracy rates than other models. This result contradicts outcomes from Danenas and Garsva (2015) and Cleofas-Sanchez, Garcia, Marques, and Senchez (2016), but corroborates, for instance, Wang, Ma, and Yang (2014), Liang, Lu, Tsai, and Shih (2016), and Cano et al. (2017). Our study supports the applicability of the expert systems by practitioners as in Heo and Yang (2014), Kim, Kang, and Kim (2015) and Xiao, Xiao, and Wang (2016). [ABSTRACT FROM AUTHOR]

• The SUDF-RS model explores the complementarity of supervised and unsupervised features for FERP. • The SUDF-RS improves RS with a feature weighting mechanism. • The SUDF-RS achieves competitive results compared with benchmarks. The supervised and unsupervised deep representation features have been adopted and verified their effective roles separately in the foreign exchange rate prediction (FERP). However, the complementarity of these two types of features for FERP remained unexplored. To address this problem, we proposed a novel method for one day ahead FERP, which improves R andom S ubspace by simultaneously considering both S upervised and U nsupervised D eep representation F eatures, namely, SUDF-RS. Feature extraction and model construction are two important stages in the SUDF-RS method. Firstly, the supervised and unsupervised deep representation features are extracted by the long short-term memory networks and deep belief networks respectively. Secondly, an improved RS method, which incorporates random forest-based feature weighting mechanism is developed to generate high-quality feature subsets. Thirdly, each feature subset is used to train the corresponding base learner and the final results are obtained by averaging the results of each base learner. Experiments on three exchange rate datasets, namely EUR/USD, GBP/USD and USD/JPY validate that the proposed SUDF-RS significantly outperforms the benchmark methods in terms of both MAPE and RMSE. [ABSTRACT FROM AUTHOR]

• Plant disease detection and diagnosing measures become a major concern in agriculture field. • Proposed HRF-MCSVM accurately classifies the diseases and rapidly improves the quality. • So avoid such destruction, plant diseases should be detected at the initial stage. • Proposed HRF-MCSVM method is compared with a few existing techniques to determine its efficiency. In recent times, plant disease detection and diagnosing measures become a major concern in the agriculture field. Earlier identification of plant disease aids the farmers to take precautionary measures thereby preventing the spread of disease to other parts of plants. Based on the severity of the diseases, the plant may undergo an attack from mild to complete destruction. So, to avoid such destruction, plant diseases should be detected at the initial stage. Therefore, this paper proposes a novel hybrid random forest Multiclass SVM (HRF-MCSVM) design for plant foliar disease detection. To improve the computation accuracy, the image features are preprocessed and segmented using Spatial Fuzzy C-Means prior to the classification process. The Plant Village dataset used consists of a total of 54,303 healthy and diseased leaf images. Finally, the performance metrics like accuracy, F-measure, specificity, sensitivity, and recall value were evaluated to determine the effectiveness of the system. The proposed HRF-MCSVM method is compared with a few existing techniques to determine its efficiency. [ABSTRACT FROM AUTHOR]

Tree-based ensemble algorithms (T E A s) have made significant advances in recent years due to their simple algorithmic design. However, when the proportion of the 'most informative' features is low, the performance of conventional T E A s degrades significantly. The primary rationale for performance degradation is that traditional algorithmic design appears to be biased toward the least informative features, and the sub-space selection procedure contains uninformative features. This paper proposes a logically randomized forest (L R F) algorithm by incorporating two different enhancements into existing T E A s. The first enhancement is made to address the issue of biasness by performing feature-level engineering. The second enhancement is the approach by which individual feature sub-spaces are selected. To derive the first enhancement, we use the graph-theoretic principle of minimal vertex cover to construct a relevant assemblage of features. Following that, the permutation-based feature importance technique is employed to calculate the 'informativeness' of the relevant features in order to infuse logical randomness into the individual trees in the forest. For the second enhancement, the stratified sampling method is used to ensure that the most informative features are present in all newly created feature sub-spaces. Consequently, individual trees are generated using the Roulette wheel-based selection (R W S) algorithm. The proposed algorithm has been evaluated on two real-world genomic data sets, ten hybrid-synthetic classification data sets, and twenty multidisciplinary benchmark data sets with varying characteristics. The experimental findings demonstrate that the L R F outperforms the existing benchmark and cutting-edge T E A s. • Proposed a novel tree-based ensemble algorithm. • Appropriate for applications with sparsely populating most informative features. • Feature engineering by graph-theoretic principal and permutation-based algorithm. • Forest generation by stratified sampling and Roulette wheel selection algorithm. • Proposed algorithm outperformed the benchmark and state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Paroxysmal atrial fibrillation (PAF) is a cardiac arrhythmia which is often challenging to diagnose because patients may be asymptomatic, and episodes are usually intermittent. In this paper we describe a method for identifying patients with a history of PAF using electrocardiogram (ECG) recordings during sinus rhythm. We analyzed, on a beat-to-beat basis, the P-waves in the sinus rhythm ECGs of 69 patients with a history of PAF and 59 healthy individuals. From each subject's P-waves, we calculated key electrocardiographic metrics, including some which are proposed here for the first time. Using means testing and feature selection methods, we discerned the features which were most useful for classification, and trained a Random Forest which identified PAF patients. Our approach achieved a 93.45% accuracy, sensitivity of 95.21%, and specificity of 91.40% using P-wave integral and novel amplitude and slope-based features which ranked highest in importance compared to other metrics from the literature. In particular, descriptive statistics of P-wave amplitudes, slopes, and integrals, were effective for identifying subjects with PAF history vs. healthy individuals. Our method has a high sensitivity and discrimination ability, with an AUC=0.9669 which is superior to others', and can thus be potentially valuable for the early identification of patients who are prone to episodes of PAF, even as part of the standard cardiological checkup that most adults undergo periodically. • Early identification of PAF during sinus rhythm with high accuracy. • P-wave signal metrics on a beat-to-beat basis are proposed. • P-wave amplitude, integral and slope features were ranked as the most important. • A RF classifier distinguishes PAF from healthy subjects with a state-of-the-art AUC. [ABSTRACT FROM AUTHOR]

Time series classification (TSC) is an important task in time series data mining and has attracted a lot of research attention. Most TSC algorithms aim to achieve high classification accuracy while reducing the computational complexity. Currently, Time Series Combination of Heterogeneous and Integrated Embedding Forest (TS-CHIEF) is considered to be one of the state-of-the-art TSC algorithms. However, compared with fast algorithms such as Time Series Forest (TSF), TS-CHIEF still has high computation cost. On the premise that the TSF algorithm is fast, we propose a new TSC algorithm, Forest based on Interval Transformation (called FIT), which takes into account both accuracy and efficiency. FIT uses cross-validation to select appropriate transformation series and corresponding interval features, and adaptively converts the interval features of each series in the process of formal training. Subsequently, the transformed feature set is combined with the random forest training FIT model. We evaluate the performance of FIT on 85 UCR time series classification datasets. The experimental results demonstrate that FIT can achieve better accuracy while maintaining high efficiency compared with the state-of-the-art methods. • A variety of series transformation and interval features to expand feature space. • Adaptive selection of transformation series and feature by cross-validation. • Forest based on Interval Transformation for time series classification. • Experiments on real datasets verify effectiveness of our proposal. [ABSTRACT FROM AUTHOR]

Recently, there has been a significant increase in malware attacks and malicious traffic. Consequently, several machine learning-based detection models have been developed to detect them. However, the detection accuracy of these models is currently evaluated using different methodologies and datasets, with some studies overstating high detection rates. The lack of a common testing approach coupled with the limited datasets used for the experiments make it challenging to compare the performances of these models to identify those that provide superior detection accuracy. A few studies have focused on benign samples and their effects on detection accuracy. The datasets used in the experiments generally consist of benign and malicious samples; hence, binary classification is used in the machine learning models. In the binary classification task, the size of a benign sample affects the classification accuracy of malicious samples, that is, it can either improve or degrade detection accuracy. In this study, we propose a novel metric for evaluating accuracy degradation by increasing benign sample size. We mainly used the FFRI dataset, which consists of 11,243 malware samples and 250,000 benign samples, and evaluated the classification accuracy with extracted strings from the malware. In addition, we obtained other malware samples that we used as supplementary to the main dataset. We increased the number of benign samples for testing by tenfold, while maintaining the malicious sample and benign training sample sizes, which resulted in a decrease of 0.293 in the F1 score. Furthermore, we confirmed that using a sufficiently sized benign training sample set mitigates accuracy degradation. Our metric can be beneficial for evaluating the benign sample size needed in binary classification and comparing accuracy. • We propose a metric for accuracy degradation by increasing benign samples. • Increasing the test benign sample size tenfold decreased the F1 score by 0.293. • Using sufficient benign training samples mitigates accuracy degradation. [ABSTRACT FROM AUTHOR]

The optimal selection of chemical features (molecular descriptors) is an essential pre-processing step for the efficient application of computational intelligence techniques in virtual screening for identification of bioactive molecules in drug discovery. The selection of molecular descriptors has key influence in the accuracy of affinity prediction. In order to improve this prediction, we examined a Random Forest (RF)-based approach to automatically select molecular descriptors of training data for ligands of kinases, nuclear hormone receptors, and other enzymes. The reduction of features to use during prediction dramatically reduces the computing time over existing approaches and consequently permits the exploration of much larger sets of experimental data. To test the validity of the method, we compared the results of our approach with the ones obtained using manual feature selection in our previous study (Perez-Sanchez, Cano, and Garcia-Rodriguez, 2014).The main novelty of this work in the field of drug discovery is the use of RF in two different ways: feature ranking and dimensionality reduction, and classification using the automatically selected feature subset. Our RF-based method outperforms classification results provided by Support Vector Machine (SVM) and Neural Networks (NN) approaches. [ABSTRACT FROM AUTHOR]

Breakthrough performances have been achieved in computer vision by utilizing deep neural networks. In this paper we propose to use random forest to classify image representations obtained by concatenating multiple layers of learned features of deep convolutional neural networks for scene classification. Specifically, we first use deep convolutional neural networks pre-trained on the large-scale image database Places to extract features from scene images. Then, we concatenate multiple layers of features of the deep neural networks as image representations. After that, we use random forest as the classifier for scene classification. Moreover, to reduce feature redundancy in image representations we derived a novel feature selection method for selecting features that are suitable for random forest classification. Extensive experiments are conducted on two benchmark datasets, i.e. MIT-Indoor and UIUC-Sports. Obtained results demonstrated the effectiveness of the proposed method. The contributions of the paper are as follows. First, by extracting multiple layers of deep neural networks, we can explore more information of image contents for determining their categories. Second, we proposed a novel feature selection method that can be used to reduce redundancy in features obtained by deep neural networks for classification based on random forest. In particular, since deep learning methods can be used to augment expert systems by having the systems essentially training themselves, and the proposed framework is general, which can be easily extended to other intelligent systems that utilize deep learning methods, the proposed method provide a potential way for improving performances of other expert and intelligent systems. [ABSTRACT FROM AUTHOR]

• Introducing combined fuzzy firefly algorithm and random forest (named FFA-RF). • Collecting a dataset by utilizing FFA for offline clustering in different applications. • Performing RF to learn the behavioral pattern of FFA in proper cluster-head selection. • Applying the trained FFA-RF model for online clustering in new unseen WSNs. Clustering in wireless sensor networks (WSNs) has proved to be one of the most efficient ways to hierarchically organize the network topology for the purposes of load-balancing and elongating the network lifetime. However, achieving optimal clustering in WSNs is an NP-hard problem, and consequently, heuristics and metaheuristics have been widely adopted. In this paper, a combined clustering technique based on a fuzzy-firefly algorithm (FFA) and random forest (RF), shortly named as: FFA-RF, is presented as an application-specific routing protocol for WSNs. Our FFA-RF protocol entails offline tuning and online routing phases: the offline phase consists of data collection using FFA, training and test of the RF, while the online phase is the actual application of the FFA-RF model to new network instances. In the offline phase, we construct an optimized fuzzy inference system based on FFA and apply it on different network topologies, to collect a comprehensive dataset. We then divide the resulting dataset into training and test sets to train and test the RF model. In the online phase, the trained RF model is used as an online clustering algorithm to estimate the fuzzy priority factor of the nodes for being cluster heads (CHs) in new network instances. To increase the generalizability of the RF for different configurations, node features as well as application features are used as inputs of the RF model. Simulation results for different network topologies demonstrate the superiority of the proposed FFA-RF protocol in prolonging the application-specific lifetime when compared against existing crisp heuristic, fuzzy heuristic, metaheuristic, and combined fuzzy-metaheuristic protocols. [ABSTRACT FROM AUTHOR]

• This paper proposes a MOMHTS optimized HRFDL architecture for diagnosing the COVID-19 and pneumonia from chest CT and X-ray images via a collaborative edge framework to offer web services in underprivileged areas. • The hyperparameters of the HRFDL classifier are optimized using the MOMHTS algorithm which minimizes the complexity of the Deep learning model that prevents it from being deployed in the edge devices. COVID-19 is a global pandemic that mostly affects patients' respiratory systems, and the only way to protect oneself against the virus at present moment is to diagnose the illness, isolate the patient, and provide immunization. In the present situation, the testing used to predict COVID-19 is inefficient and results in more false positives. This difficulty can be solved by developing a remote medical decision support system that detects illness using CT scans or X-ray images with less manual interaction and is less prone to errors. The state-of-art techniques mainly used complex deep learning architectures which are not quite effective when deployed in resource-constrained edge devices. To overcome this problem, a Multi-Objective Modified Heat Transfer Search (MOMHTS) optimized Hybrid Random Forest Deep learning (HRFDL) classifier is proposed in this paper. The MOMHTS algorithm mainly optimizes the deep learning model in the HRFDL architecture by optimizing the hyperparameters associated with it to support the resource-constrained edge devices. To evaluate the efficiency of this technique, extensive experimentation is conducted on two real-time datasets namely the COVID19 lung CT scan dataset and the Chest X-ray images (Pneumonia) datasets. The proposed methodology mainly offers increased speed for communication between the IoT devices and COVID-19 detection via the MOMHTS optimized HRFDL classifier is modified to support the resources which can only support minimal computation and handle minimum storage. The proposed methodology offers an accuracy of 99% for both the COVID19 lung CT scan dataset and the Chest X-ray images (Pneumonia) datasets with minimal computational time, cost, and storage. Based on the simulation outcomes, we can conclude that the proposed methodology is an appropriate fit for edge computing detection to identify the COVID19 and pneumonia with higher detection accuracy. [ABSTRACT FROM AUTHOR]

The popularity of the online peer-to-peer (P2P) lending platform in the financial supply chain (FSC) has grown tremendously in the past few years. However, it is pretty challenging to develop an efficient financial supply chain for Micro, Small, and Medium-Sized Enterprises (MSMEs) by providing credits and services. In this research, an FSC model analyzes by emphasizing the retailer and supplier relationship. The retailer predicts the market demand to maximize profit, and accordingly, orders to the suppliers and suppliers utilize the production capabilities in limited monetary value. In such a case, the supplier also needs to apply for loans to meet the production commitment. With an intelligent lending platform, applicants can get loans quickly and more conveniently from the lenders. Still, identifying defaulter borrowers is a difficult task on the peer-to-peer lending platform. The main focus of the lenders is to maximize profit and minimize risk by giving the loan to non-defaulter suppliers. The present study describes the importance of the financial parameters (λ) of the retailers and suppliers (MSMEs) and their impact on the related supply chain. We provide the proof of dependency of the financial parameter on the cost of debt and the ratio of the debt-to-equity by developing two propositions. We propose an innovative k -Random Boosting Classifiers (k -RBC) algorithm for identifying potential good and bad borrowers to capture this case. The time complexity of the k -RBC algorithm is O (n 2 l o g (n)). The results obtained from the study show a significant improvement in comparison to the outputs from existing approaches on the same datasets. Our algorithm gives 90% accuracy to identify potential good borrowers, whereas existing algorithms achieve up to 87% accuracy. Furthermore, the importance of the borrowers' features and its impact are analyzed of the lending platform in FSC. • Defines the importance of financial parameters and its impact on supply chain. • Provides the proof of dependency of the financial parameter on the cost of debt. • Prove dependency of the financial parameter on the ratio of the debt-to-equity. • Proposes the k-Random Boosting Classification model to identify defaulters. • Identifies crucial features to enhance the probability of getting loans approval. [ABSTRACT FROM AUTHOR]

• Select the most significant features of Polycystic Ovary Syndrome patients. • Analyzed different machine learning procedures on the PCOS dataset. • Find the optimum set of parameters for the Random Forest classifier. • Achieved an overall classification accuracy of 93.25%. Polycystic Ovary Syndrome (PCOS) is a hormonal disorder that affects a large percentage of women of reproductive age. PCOS causes imbalanced or delayed menstrual cycles and produces high levels of the male hormone. The ovaries may create a significant number of little fluid-filled sacs (follicles) yet fail to discharge eggs regularly. The actual cause of PCOS is uncertain. However, early exposure and curing, as well as weight loss, may lower the threat of long-term complications. This study focuses on PCOS diagnosis based on a clinical dataset supplied by Kottarathil, accessible via its Kaggle repository. Non -invasive screening parameters are used to evaluate a range of machine learning approaches for screening PCOS patients without the use of invasive diagnostics. According to the findings of the experiments, the Random Forest (RF) method outperforms the other prominent machine learning algorithms with an accuracy of 93.25%. Further, the out-of-bag (OOB) error is utilized for assessing the prediction performance of RF. [ABSTRACT FROM AUTHOR]

• A novel model is proposed to predict the severity of Parkinson's disease. • More efficient frequency features are obtained by graph wavelet transform. • The attentional mechanism weights decision tree models in the random forest. • We achieve better prediction performance compared to the state-of-the-art methods. The progress prediction of Parkinson's disease (PD) is one of the most important issues in early diagnosis of PD. Many researches have been conducted in this field, however, most existing methods focus on the selection of baseline features and regressors to reduce prediction errors. Different from the previous studies, the main goal of this paper is to obtain more effective features by feature transformation of baseline features to improve the prediction performance. Therefore, this paper proposes a prediction model based on graph wavelet transform (GWT) and attention weighted random forest (RF). Firstly, a clustering algorithm is adopted to reduce the prediction error of the model. Next, a multi-scale analysis of the feature vectors by GWT is conducted to yield a frequency feature representation that is more relevant to the target value. Finally, the frequency features are input into the attention weighted RF to predict the severity of PD, allowing the results of decision trees with better predictive performance in the RF to be highlighted while reducing the risk of overfitting. The effectiveness of the method is evaluated on the Parkinson's telemonitoring dataset collected by the University of Oxford. The experimental results show that the mean absolute error and root mean squared error of the proposed method for predicting PD severity (motor- and total-UPDRS) are 1.53, 2.13 and 1.91, 2.70, respectively. Compared with the quoted optimal method, the errors are reduced by 7.27%, 4.05% and 5.45%, 1.10%, respectively. This indicates that the proposed method has better prediction performance. [ABSTRACT FROM AUTHOR]

This paper presents a novel approach to modelling visual distraction of bicyclists. A unique bicycle simulator equipped with sensors capable of capturing the behaviour of the bicyclist is presented. While cycling two similar scenario routes, once while simultaneously interacting with an electronic device and once without any electronic device, statistics of the measured speed, head movements, steering angle and bicycle road position along with questionnaire data are captured. These variables are used to model the self-assessed distraction level of the bicyclist. Data mining techniques based on random forests, support vector machines and neural networks are evaluated for the modelling task. Out of the total 71 measured variables a variable selection procedure based on random forests is able to select a fraction of those and consequently improving the modelling performance. By combining the random forest-based variable selection and support vector machine-based modelling technique the best overall performance is achieved. The method shows that with a few observable variables it is possible to use machine learning to model, and thus predict, the distraction level of a bicyclist. [ABSTRACT FROM AUTHOR]

Topic of this study is exploration and fusion of non-invasive measurements for an accurate detection of pathological larynx. Measurements for human subject encompass answers to items of a specific survey and information extracted by the openSMILE toolkit from several audio recordings of sustained phonation (vowel /a/). Clinical diagnosis, assigned by medical specialist, is a target attribute distinguishing subject as healthy or pathological. Random forest (RF) is used here as a base-learner and also as a meta-learner for decision-level fusion. 5 RF classifiers, built separately on 3 variants of audio recording data (raw and after two types of voice activity detection) and 2 variants of questionnaire (with 9 and 26 questions) data, are fused selectively by finding out the best combination of all possible. Before fusion, due to presence of missing values in query modalities, several imputation techniques were evaluated besides the complete-case analysis by listwise deletion. Out-of-bag equal error rate (EER) was found to be higher for audio data and lower for query, but each variant was outperformed by the decision-level fusion. Fusion after listwise deletion provided EER of 4.84%, meanwhile imputation was found to improve detection slightly and helped to achieve EER of 4.55%. Variable importance, as permutation-based mean decrease in RF accuracy, was reported for query and audio data. Finally, regarding the noteworthy performance of the query data, 22 association rules (11 healthy + 11 pathological) using 17 questions were obtained for comprehensible insights. [ABSTRACT FROM AUTHOR]

Time series classification is an important topic in data mining. Time series classification methods have been studied by many researchers. A feature-weighted classification method is proposed based on complex network. There are four stages in the proposed classification algorithm. First, we obtain visible points and edge weights by applying an improved weighted visibility graph algorithm to transform the original time series into weighted complex networks. Next, weighted features are calculated based on visible points and weights, and unable-weighted features are obtained by utilizing NetworkX. After that, we combine two types of features to obtain a total features matrix. We then reconsider the weight of each independent feature by utilizing the importance function of random forest to measure the relative importance of features and construct a importance weighted features matrix. Finally, we utilize a traditional classifier called random forest to cluster the weighted features matrix and generate classification results. Simultaneously, a novel feature weight calculation method is applied during the classification process. We compare the proposed method to other classification methods and the results indicate that the proposed method can improve classification accuracy for time series datasets. • The complex network improves the accuracy of classification results. • The features of complex network reflect the dynamic information. • This method can select the features to better reflect the complex network structure. • The weighted features can improve the classification effect. [ABSTRACT FROM AUTHOR]