Feature Selection Method Research Articles

A novel method for optimizing epilepsy detection features through multi-domain feature fusion and selection.

The methods used to detect epileptic seizures using electroencephalogram (EEG) signals suffer from poor accuracy in feature selection and high redundancy. This problem is addressed through the use of a novel multi-domain feature fusion and selection method (PMPSO). Discrete Wavelet Transforms (DWT) and Welch are used initially to extract features from different domains, including frequency domain, time-frequency domain, and non-linear domain. The first step in the detection process is to extract important features from different domains, such as frequency domain, time-frequency domain, and non-linear domain, using methods such as Discrete Wavelet Transform (DWT) and Welch. To extract features strongly correlated with epileptic classification detection, an improved particle swarm optimization (PSO) algorithm and Pearson correlation analysis are combined. Finally, Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest (RF) and XGBoost classifiers are used to construct epileptic seizure detection models based on the optimized detection features. According to experimental results, the proposed method achieves 99.32% accuracy, 99.64% specificity, 99.29% sensitivity, and 99.32% score, respectively. The detection performance of the three classifiers is compared using 10-fold cross-validation. Surpassing other methods in detection accuracy. Consequently, this optimized method for epilepsy seizure detection enhances the diagnostic accuracy of epilepsy seizures.

Feature selection based on consistent granulation

Classification learning from large-scale data is a pivotal and complex task in machine learning. Enhancing classifier performance can be achieved through effective feature selection and construction. However, current feature selection methods often encounter difficulties with large-scale data. Considering granulation is essential for developing accurate and efficient models, we firstly introduce a novel granulation technique, termed consistent granulation in this study. Furthermore, an innovative feature construction method leveraging consistent granulation is proposed to enhance computational speed, scalability, and classification performance of feature selection. Consequently, both time and space complexities are reduced to linear levels relative to sample sizes. Consistent granulation also allows for an accurate representation of the data's topological structure and reduces the parameter set from the interval [0,1] to a finite set, simplifying the search for optimal parameters. Experimental results indicate that our proposed algorithm, FCG, surpasses four classical feature selection algorithms—FARNeMF, HANDI, GBNRS, and RMDPS—in handling datasets with millions of samples and tens of thousands of dimensions. Specifically, on the first fourteen datasets, FCG decreases the average runtime by times of 1029.30 (FARNeMF), 703.54 (HANDI), 1253.78 (GBNRS), and 3.08 (RMDPS), while enhancing average classification accuracy by 3.24% (FARNeMF), 2.74% (HANDI), 5.77% (GBNRS), and 15.5% (RMDPS).

A hybrid feature selection method for text classification using a feature-correlation-based genetic algorithm

A hybrid feature selection method for text classification using a feature-correlation-based genetic algorithm

Comparative Analysis of Feature Selection Methods with XGBoost for Malware Detection on the Drebin Dataset

Malware, or malicious software, continues to evolve alongside increasing cyberattacks targeting individual devices and critical infrastructure. Traditional detection methods, such as signature-based detection, are often ineffective against new or polymorphic malware. Therefore, advanced malware detection methods are increasingly needed to counter these evolving threats. This study aims to compare the performance of various feature selection methods combined with the XGBoost algorithm for malware detection using the Drebin dataset, and to identify the best feature selection method to enhance accuracy and efficiency. The experimental results show that XGBoost with the Information Gain method achieves the highest accuracy of 98.7%, with faster training times than other methods like Chi-Squared and ANOVA, which each achieved an accuracy of 98.3%. Information Gain yielded the best performance in accuracy and training time efficiency, while Chi-Squared and ANOVA offered competitive but slightly lower results. This study highlights that appropriate feature selection within machine learning algorithms can significantly improve malware detection accuracy, potentially aiding in real-world cybersecurity applications to prevent harmful cyberattacks.

Development of compact smart bearing and novel hybrid feature assessment for weak defect identification

ABSTRACT In this study, a compact smart bearing with self-sensing and condition monitoring function is proposed. A traditional bearing embedded in a piezoelectric transducer ring with a segmented electrode design is prototyped and tested under different working conditions. Its compact design ensures compatibility for integration with different systems. The segmented electrodes enable possible multidirectional data acquisition and piezoelectric energy generation, suggesting potential integration with Internet of Things (IoT) platforms. Utilizing the high sensitivity, fast response and high signal resolution of the piezoelectric material in conjunction with a novel design that enables close contact between the piezoelectric transducer and the bearing, the smart bearing demonstrates effective performance in detecting weak bearing defects signals. A novel feature characterising method is proposed, and a hybridised feature selection method is employed for reducing the dimension of feature subsets and ensuring defect identification accuracy. A classification model for the identification of defects is developed based on a Long Short-Term Memory (LSTM) network. The performance of the smart bearing and the method for identifying the defects are evaluated through experiments to demonstrate the potential for practical applications. A preliminary experiment for energy harvesting using the smart bearing has been conducted, and it proves the potential to sustain power.

Multi-Label Feature Selection with Feature-Label Subgraph Association and Graph Representation Learning.

In multi-label data, a sample is associated with multiple labels at the same time, and the computational complexity is manifested in the high-dimensional feature space as well as the interdependence and unbalanced distribution of labels, which leads to challenges regarding feature selection. As a result, a multi-label feature selection method based on feature-label subgraph association with graph representation learning (SAGRL) is proposed to represent the complex correlations of features and labels, especially the relationships between features and labels. Specifically, features and labels are mapped to nodes in the graph structure, and the connections between nodes are established to form feature and label sets, respectively, which increase intra-class correlation and decrease inter-class correlation. Further, feature-label subgraphs are constructed by feature and label sets to provide abundant feature combinations. The relationship between each subgraph is adjusted by graph representation learning, the crucial features in different label sets are selected, and the optimal feature subset is obtained by ranking. Experimental studies on 11 datasets show the superior performance of the proposed method with six evaluation metrics over some state-of-the-art multi-label feature selection methods.

Prediction of esophageal fistula in radiotherapy/chemoradiotherapy for patients with advanced esophageal cancer by a clinical-deep learning radiomics model : Prediction of esophageal fistula in radiotherapy/chemoradiotherapy patients.

Esophageal fistula (EF), a rare and potentially fatal complication, can be better managed with predictive models for personalized treatment plans in esophageal cancers. We aim to develop a clinical-deep learning radiomics model for effectively predicting the occurrence of EF. The study involved esophageal cancer patients undergoing radiotherapy or chemoradiotherapy. Arterial phase enhanced CT images were used to extract handcrafted and deep learning radiomic features. Along with clinical information, a 3-step feature selection method (statistical tests, Least Absolute Shrinkage and Selection Operator, and Recursive Feature Elimination) was used to identify five feature sets in training cohort for constructing random forest EF prediction models. Model performance was compared and validated in both retrospective and prospective test cohorts. One hundred seventy five patients (122 in training and 53 in test cohort)were retrospectively collected from April 2018 to June 2022. An additional 27 patients were enrolled as a prospective test cohort from June 2022 to December 2023. Post-selection in the training cohort, five feature sets were used for model construction: clinical, handcrafted radiomic, deep learning radiomic, clinical-handcrafted radiomic, and clinical-deep learning radiomic. The clinical-deep learning radiomic model excelled with AUC of 0.89 (95% Confidence Interval: 0.83-0.95) in the training cohort, 0.81 (0.65-0.94) in the test cohort, and 0.85 (0.71-0.97) in the prospective test cohort. Brier-score and calibration curve analyses validated its predictive ability. The clinical-deep learning radiomic model can effectively predict EF in patients with advanced esophageal cancer undergoing radiotherapy or chemoradiotherapy.

Laser-Induced Breakdown Spectroscopy and a Convolutional Neural Network Model for Predicting Total Iron Content in Iron Ores.

Laser-induced breakdown spectroscopy (LIBS) is a rapid method for detecting total iron (TFe) content in iron ores. However, accuracy and measurement error of univariate regression analysis in LIBS are limited due to factors such as laser energy fluctuations and spectral interference. To address this, multiple regression analysis and feature selection/extraction are needed to reduce redundant information, decrease the correlation between variables, and quantify the TFe content of iron ores accurately. Overall, 339 batches of iron ore samples from five countries were obtained from the ports of China during the discharging, and 2034 representative spectra were collected. A convolutional neural network (CNN) model for total iron content prediction in iron ores is established. The performance of variable importance random forest (VI-RF), variable importance back propagation artificial neural network (VI-BP-ANN), and CNN-assisted LIBS in predicting the TFe content of iron ores was compared. Coefficient of determination (R2), root mean square error (RMSE), mean relative error (MRE), and modeling time were selected for model evaluation. The result shows that variable importance significantly enhances the quantitative accuracy and reduces modeling time compared to traditional BP-ANN and RF models. Moreover, the CNN model outperformed manual feature selection methods (VI-BP-ANN and VI-RF), exhibiting the shortest modeling time, highest R2, lowest RMSE, and MRE. CNN model's unique characteristics, such as weight sharing and local connection, make it well suited for analyzing high-dimensional LIBS data in multivariate regression analysis. Our approach demonstrates the effectiveness of machine learning and deep learning approaches in improving the accuracy of LIBS for TFe content prediction in iron ores. CNN-assisted LIBS method holds great potential for practical applications in the mining industry.

Optimizing GEDI Canopy Height Estimation and Analyzing Error Impact Factors Under Highly Complex Terrain and High-Density Vegetation Conditions

The Global Ecosystem Dynamics Investigation (GEDI) system provides essential data for estimating forest canopy height on a global scale. However, factors such as complex topography and dense canopy can significantly reduce the accuracy of GEDI canopy height estimations. We selected the South Taihang region of Henan Province, China, as our study area and proposed an optimization framework to improve GEDI canopy height estimation accuracy. This framework includes correcting geolocation errors in GEDI footprints, screening and analyzing features that affect estimation errors, and combining two regression models with feature selection methods. Our findings reveal a geolocation error of 4 to 6 m in GEDI footprints at the orbital scale, along with an overestimation of GEDI canopy height in the South Taihang region. Relative height (RH), waveform characteristics, topographic features, and canopy cover significantly influenced the estimation error. Some studies have suggested that GEDI canopy height estimates for areas with high canopy cover lead to underestimation, However, our study found that accuracy increased with higher canopy cover in complex terrain and dense vegetation. The model’s performance improved significantly after incorporating the canopy cover parameter into the optimization model. Overall, the R2 of the best-optimized model was improved from 0.06 to 0.61, the RMSE was decreased from 8.73 m to 2.23 m, and the rRMSE decreased from 65% to 17%, resulting in an accuracy improvement of 74.45%. In general, this study reveals the factors affecting the accuracy of GEDI canopy height estimation in areas with complex terrain and dense vegetation cover, on the premise of minimizing GEDI geolocation errors. Employing the proposed optimization framework significantly enhanced the accuracy of GEDI canopy height estimates. This study also highlighted the crucial role of canopy cover in improving the precision of GEDI canopy height estimation, providing an effective approach for forest monitoring in such regions and vegetation conditions. Future studies should further improve the classification of tree species and expand the diversity of sample tree species to test the accuracy of canopy height estimated by GEDI in different forest structures, consider the distortion of optical remote sensing images caused by rugged terrain, and further mine the information in GEDI waveforms so as to enhance the applicability of the optimization framework in more diverse forest environments.

Stress recognition identifying relevant facial action units through explainable artificial intelligence and machine learning

Background and objective: Facial cues and expressions constitute a component of bodily responses that provide useful information about one’s stress levels. According to the Facial Action Coding System, they can be modelled consistently in terms of fundamental facial muscle movements, called facial Action Units (AUs). This article investigates automatic acute stress recognition based on AUs using conventional Machine and Deep Learning techniques. Methods: We created a new experimental dataset containing 58 participants performing 4 experimental phases and 11 stress and non-stress tasks in which the proposed system performs automatic facial AUs recognition. A computational feature selection method was employed to select a robust relevant AU combinations subset, which integrated with conventional Machine Learning and Deep Learning methods using the Layer-Wise Relevance Propagation algorithm to assess and model the implication of AUs under acute stress conditions. Ordinal modelling was used following the pairwise transformation to establish a common reference based on the personalized values of each participant. Results: The results indicate that, under acute stress conditions, participants’ faces presented significantly more AUs and with greater intensity compared to neutral conditions. The most relevant combination of AUs to each stress type was computationally identified, ranked and selected. The mean yielded classification accuracy of stress condition versus neutral achieved across all experimental tasks was greater than 93%. Conclusions: There are specific combinations of AUs that are relevant to the stress conditions of each experimental phase leading in each case to better neutral and stress separability.

An Ensemble Machine Learning With Feature Selection Methods for Detecting Phishing Attacks

An Ensemble Machine Learning With Feature Selection Methods for Detecting Phishing Attacks

Impact of Feature Selection Techniques on the Performance of Machine Learning Models for Depression Detection Using EEG Data

Major depressive disorder (MDD) poses a significant challenge in mental healthcare due to difficulties in accurate diagnosis and timely identification. This study explores the potential of machine learning models trained on EEG-based features for depression detection. Six models and six feature selection techniques were compared, highlighting the crucial role of feature selection in enhancing classifier performance. This study investigates the six feature selection methods: Elastic Net, Mutual Information (MI), Chi-Square, Forward Feature Selection with Stochastic Gradient Descent (FFS-SGD), Support Vector Machine-based Recursive Feature Elimination (SVM-RFE), and Minimal-Redundancy-Maximal-Relevance (mRMR). These methods were combined with six diverse classifiers: Logistic Regression, Support Vector Machine (SVM), Random Forest, Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM). The results demonstrate the substantial impact of feature selection on model performance. SVM-RFE with SVM achieved the highest accuracy (93.54%) and F1 score (95.29%), followed by Logistic Regression with an accuracy of 92.86% and F1 score of 94.84%. Elastic Net also delivered strong results, with SVM and Logistic Regression both achieving 90.47% accuracy. Other feature selection methods yielded lower performance, emphasizing the importance of selecting appropriate feature selection and machine learning algorithms. These findings suggest that careful selection and application of feature selection techniques can significantly enhance the accuracy of EEG-based depression detection.

Causality-Driven Feature Selection for Calibrating Low-Cost Airborne Particulate Sensors Using Machine Learning.

With escalating global environmental challenges and worsening air quality, there is an urgent need for enhanced environmental monitoring capabilities. Low-cost sensor networks are emerging as a vital solution, enabling widespread and affordable deployment at fine spatial resolutions. In this context, machine learning for the calibration of low-cost sensors is particularly valuable. However, traditional machine learning models often lack interpretability and generalizability when applied to complex, dynamic environmental data. To address this, we propose a causal feature selection approach based on convergent cross mapping within the machine learning pipeline to build more robustly calibrated sensor networks. This approach is applied in the calibration of a low-cost optical particle counter OPC-N3, effectively reproducing the measurements of PM1 and PM2.5 as recorded by research-grade spectrometers. We evaluated the predictive performance and generalizability of these causally optimized models, observing improvements in both while reducing the number of input features, thus adhering to the Occam's razor principle. For the PM1 calibration model, the proposed feature selection reduced the mean squared error on the test set by 43.2% compared to the model with all input features, while the SHAP value-based selection only achieved a reduction of 29.6%. Similarly, for the PM2.5 model, the proposed feature selection led to a 33.2% reduction in the mean squared error, outperforming the 30.2% reduction achieved by the SHAP value-based selection. By integrating sensors with advanced machine learning techniques, this approach advances urban air quality monitoring, fostering a deeper scientific understanding of microenvironments. Beyond the current test cases, this feature selection method holds potential for broader applications in other environmental monitoring applications, contributing to the development of interpretable and robust environmental models.

AI tool for predicting MGMT methylation in glioblastoma for clinical decision support in resource limited settings

Glioblastoma is an aggressive brain cancer with a poor prognosis. The O6-methylguanine-DNA methyltransferase (MGMT) gene methylation status is crucial for treatment stratification, yet economic constraints often limit access. This study aims to develop an artificial intelligence (AI) framework for predicting MGMT methylation. Diagnostic magnetic resonance (MR) images in public repositories were used for training. The algorithm created was validated in data from a single institution. All images were segmented according to widely used guidelines for radiotherapy planning and combined with clinical evaluations from neuroradiology experts. Radiomic features and clinical impressions were extracted, tabulated, and used for modeling. Feature selection methods were used to identify relevant phenotypes. A total of 100 patients were used for training and 46 for validation. A total of 343 features were extracted. Eight feature selection methods produced seven independent predictive frameworks. The top-performing ML model was a model post-Least Absolute Shrinkage and Selection Operator (LASSO) feature selection reaching accuracy (ACC) of 0.82, an area under the curve (AUC) of 0.81, a recall of 0.75, and a precision of 0.75. This study demonstrates that integrating clinical and radiotherapy-derived AI-driven phenotypes can predict MGMT methylation. The framework addresses constraints that limit molecular diagnosis access.

Computer-aided diagnosis for China-Japan Friendship Hospital classification of necrotic femurs using statistical shape and appearance model based on CT scans.

The purpose of this study is to quantify the three-dimensional (3D) structural morphology, bone mineral density (BMD) distribution, and mechanical properties of different China-Japan Friendship Hospital (CJFH) classification types and assist clinicians in classifying necrotic femurs accurately. In this study, 41 cases were classified as types L2 and L3 based on CT images. Then, 3D Statistical Shape and Appearance Models (SSM and SAM) were established, and 80 principal component (PC) modes were extracted from the SSM and SAM as the candidate features. The bone strength of each case was also calculated as the candidate feature using finite element analysis (FEA). Support vector machine (SVM) and Extreme Gradient Boosting (XGBoost) were used to establish 10 machine learning models. Feature selection methods were used to screen the candidate features. The performance of each model was evaluated based on sensitivity, specificity, accuracy, and the area under the receiver operating characteristic (ROC) curve. This resulted in a SVM model for CJFH classification with the performance: accuracy of 87.5%, sensitivity of 85.0%, specificity of 76.0%, and AUC of 94.2%. This study provided effective machine learning models for assisting in diagnosing CJFH types, increasing the objectivity of the diagnosis. They may have great potential for application in clinical assessments of CJFH classification.

Robust remote detection of depressive tendency based on keystroke dynamics and behavioural characteristics

Depressive Disorder (DD) is a leading cause of disability worldwide. Screening tools for detecting DD symptoms are essential for monitoring and efficient managing. Remarkably, individuals’ kinetic activities, including their interaction with touchscreen smartphones, can be a proxy for their mental status. Therefore, studying these typing patterns can assist in developing passive screening tools for detecting even the early stage of DD, i.e., the depressive tendency (DT). Here we extend a previous study by exploring different machine learning models with various feature engineering approaches to detect the subjects’ DT, as indicated by the self-administered Patient Health Questionnaire-9 (PHQ-9) score, via keystroke digital biomarkers. The keystroke timing sequences were unobtrusively collected from 24 subjects during routine interaction with touchscreen smartphones, resulting in 23,264 typing sessions. The proposed framework was investigated under two keystroke feature combinations—hold-time and flight-time variables—and validated using nested cross-validation scheme. Different feature selection (FS) techniques were employed to select informative features from the keystroke sequences. The best-performing gradient boosting classifier with features selected by the mutual information FS method achieved an improved Area Under Curve (AUC) of 0.98 [95% confidence interval: 0.91-1.00]. The proposed DT pipeline, which surpasses the state-of-the-art models, could effectively capture DT, considering users’ behavioural characteristics. This would potentially provide users with information regarding the evolution of their mental health, simultaneously contributing to improving digital tools for objectively screening mental disorders in-the-wild.

Novel statistically equivalent signature-based hybrid feature selection and ensemble deep learning LSTM and GRU for chronic kidney disease classification

Chronic kidney disease (CKD) involves numerous variables, but only a few significantly impact the classification task. The statistically equivalent signature (SES) method, inspired by constraint-based learning of Bayesian networks, is employed to identify essential features in CKD. Unlike conventional feature selection methods, which typically focus on a single set of features with the highest predictive potential, the SES method can identify multiple predictive feature subsets with similar performance. However, most feature selection (FS) classifiers perform suboptimally with strongly correlated data. The FS approach faces challenges in identifying crucial features and selecting the most effective classifier, particularly in high-dimensional data. This study proposes using the Least Absolute Shrinkage and Selection Operator (LASSO) in conjunction with the SES method for feature selection in CKD identification. Following this, an ensemble deep-learning model combining long short-term memory (LSTM) and gated recurrent unit (GRU) networks is proposed for CKD classification. The features selected by the hybrid feature selection method are fed into the ensemble deep-learning model. The model’s performance is evaluated using accuracy, precision, recall, and F1 score metrics. The experimental results are compared with individual classifiers, including decision tree (DT), Random Forest (RF), logistic regression (LR), and support vector machine (SVM). The findings indicate a 2% improvement in classification accuracy when using the proposed hybrid feature selection method combined with the LSTM and GRU ensemble deep-learning model. Further analysis reveals that certain features, such as HEMO, POT, bacteria, and coronary artery disease, contribute minimally to the classification task. Future research could explore additional feature selection methods, including dynamic feature selection that adapts to evolving datasets and incorporates clinical knowledge to enhance CKD classification accuracy further.

A knowledge empowered graph learning feature selection method based on variation propagation effect representation and analysis for human-centric manufacturing systems

As the construction goals of manufacturing systems shift from Industry 4.0 to Industry 5.0, the development goals of manufacturing systems have also shifted from technology driven to human-centric. Feature Selection (FS) of manufacturing systems plays an important role in helping human control and upgrade manufacturing system quality. However, due to the propagation of variations in manufacturing systems, traditional FS methods cannot accurately select the key features that have an important disturbance to quality indicator. In this paper, a process knowledge-enabled FS method is proposed to obtain key features of quality indictor (QI) that would cause bias of prediction model accuracy. Firstly, a process representation learning algorithm embedded in the process knowledge graph is proposed, which represents the manufacturing variation propagation process during the manufacturing process. Secondly, a feature importance evaluation algorithm based on disturbance intensity judgment rules is proposed to evaluate the importance of process features. Finally, the top k important features are selected as key features based on feature importance ranking. Experimental results demonstrate the proposed method outperforms state-of-art and classic methods in four evaluated indicators. Moreover, the four indicators of MSE, MAE, RMSE, and R 2 have respectively increased by 18%, 16%, 18%, and 33% with 7 features compared to the cutting-edge method.

Comparison of automatic and physiologically-based feature selection methods for classifying physiological stress using heart rate and pulse rate variability indices.

This study evaluates the effectiveness of four machine Learning algorithms in classifying physiological stress using Heart Rate Variability (HRV) and Pulse Rate Variability (PRV) time series, comparing an automatic feature selection based on Akaike's criterion to a physiologically-based feature selection approach. Linear Discriminant Analysis, Support Vector Machines, K-Nearest Neighbors and Random Forest were applied on ten HRV and PRV indices from time, frequency and information domains, selected with the two feature selection approaches. Data were collected from 127 healthy individuals during different stress conditions (rest, postural and mental stress). Our results highlight that, while specific stress classification is feasible, distinguishing between postural and mental stress remains challenging. The used classifiers exhibited similar performance, with automatic Akaike Information Criterion-based feature selection proving overall better than the physiology-driven approach. Additionally, PRV-based features performed comparably to HRV-based ones, indicating their potential in outpatient monitoring using wearable devices. The obtained findings help to determine the most relevant HRV/PRV features for stress classification, potentially useful to highlight different physiological mechanisms involved during both challenges accompanied by a shift in the sympathovagal balance. The proposed approach may have implications for advancing stress assessment methodologies in clinical settings and real-world contexts for well-being evaluation.

Multiple feature selection based on an optimization strategy for causal analysis of health data

PurposeRecent advancements in information technology and wearable devices have revolutionized healthcare through health data analysis. Identifying significant relationships in complex health data enhances healthcare and public health strategies. In health analytics, causal graphs are important for investigating the relationships among health features. However, they face challenges owing to the large number of features, complexity, and computational demands. Feature selection methods are useful for addressing these challenges. In this paper, we present a framework for multiple feature selection based on an optimization strategy for causal analysis of health data.MethodsWe select multiple health features based on an optimization strategy. First, we define a Weighted Total Score (WTS) index to assess the feature importance after the combination of different feature selection methods. To explore an optimal set of weights for each method, we design a multiple feature selection algorithm integrated with the greedy algorithm. The features are then ranked according to their WTS, enabling selection of the most important ones. After that, causal graphs are constructed based on the selected features, and the statistical significance of the paths is assessed. Furthermore, evaluation experiments are conducted on an experiment dataset collected for this study and an open dataset for diabetes.ResultsThe results demonstrate that our approach outperforms baseline models by reducing the number of features while improving model performance. Moreover, the statistical significance of the relationships between features uncovered through causal graphs is validated for both datasets.ConclusionBy using the proposed framework for multiple feature selection based on an optimization strategy for causal analysis, the number of features is reduced and the causal relationships are uncovered and validated.