From fractions to fragments: Policy to practice through AI-driven multiscale spatial planning for groundwater nitrate management.

High-dimensional datasets are commonly encountered in real-world machine learning applications and often degrade classification performance due to redundant and irrelevant features. In addition, the presence of excessive features increases computational complexity and processing time. Feature selection is therefore a crucial preprocessing step to improve model accuracy and efficiency. This study proposes a hybrid feature selection approach called Intersection Filtering based on Recursive Feature Elimination with Cross-Validation (IF-RFECV), which integrates wrapper-based and filter-based strategies to obtain a stable and optimal subset of features. The proposed method first applies Recursive Feature Elimination with Cross-Validation (RFECV) using multiple classification models to rank and select relevant features. Subsequently, an intersection filtering mechanism is employed to identify features that are consistently selected across different RFECV-based models, thereby reducing model-dependent bias and improving feature robustness. The effectiveness of IF-RFECV is evaluated using four benchmark datasets with varying dimensionality obtained from the KEEL and UCI repositories. Several classification algorithms, including Gradient Boosting, K-Nearest Neighbor, Naïve Bayes, Decision Tree, Random Forest, and Support Vector Machine, are used to assess model performance. Experimental results demonstrate that IF-RFECV produces a more compact feature subset compared to conventional RFECV while achieving superior performance in terms of accuracy, precision, recall, and F1-score on most datasets, particularly those with higher dimensionality. Although IF-RFECV requires slightly higher computational time due to its two-stage process, the performance gains and improved generalization justify this trade-off. These findings indicate that IF-RFECV is an effective and robust feature selection technique for high-dimensional classification problems.

The rapid increase in cardiovascular diseases has necessitated the development of intelligent, scalable, and realtime healthcare solutions capable of early diagnosis and prevention. Smart healthcare systems integrating the Internet of Things (IoT) and Machine Learning (ML) have emerged as transformative technologies that enable continuous monitoring, data-driven decision-making, and predictive analytics in clinical environments. This study presents a comprehensive framework for heart disease prediction using IoT-enabled sensing devices and advanced machine learning algorithms, emphasizing real-world applicability and algorithmic development. The proposed system leverages wearable and embedded sensors to collect physiological parameters such as heart rate, blood pressure, and electrocardiogram signals, which are transmitted through cloud-based architectures for preprocessing and analysis. Machine learning models, including supervised and ensemble approaches, are developed to identify patterns and predict cardiovascular risk with high accuracy. The study further explores optimization strategies, feature selection techniques, and model interpretability to enhance predictive performance and clinical reliability. Real-world implementation scenarios are analyzed to demonstrate the feasibility of integrating such systems into modern healthcare infrastructures, including remote patient monitoring and telemedicine platforms. Additionally, the research highlights the challenges associated with data quality, privacy, interoperability, and scalability while proposing solutions for robust deployment. The findings indicate that IoT and ML-based healthcare systems significantly improve early diagnosis, reduce mortality rates, and support personalized treatment strategies. This research contributes to the advancement of intelligent healthcare by bridging the gap between theoretical models and practical applications in heart disease prediction.

Personalized weight loss management through wearable devices and artificial intelligence.

Due to increasing complexity and frequency of credit card fraud, there is a critical need for highly accurate and efficient detection systems. This study proposes an enhanced fraud detection framework that combines ensemble learning with feature importance techniques to improve performance. It uses six powerful boosting algorithms — AdaBoost, XGBoost, GBM, LightGBM, CatBoost, and LogitBoost — as base models, which are then merged using a stacked ensemble method to boost prediction accuracy. To ensure model efficiency and interpretability, feature selection techniques such as recursive feature elimination, tree-based importance, mutual information classification, and ANOVA F-test are applied, with the ANOVA method prioritized in the final model. When evaluated on a benchmark dataset, the proposed system achieved exceptional results: accuracy, precision, recall, and F-measure of 99.97 %. This demonstrates the effectiveness of stacked ensembles in combining the strengths of individual models while minimizing errors. The feature selection process also improves computational efficiency by focusing on the most relevant features.

Not all data features are crucial for uncovering hidden knowledge within various datasets, making the reduction of their dimensional attributes a significant area of interest. In this work, a new meta-heuristic algorithm IGWO-RF which is a combination of an improved gray wolf optimization (GWO) algorithm and random forest (RF) is suggested for feature selection problems. In the improved GWO, a nonlinear variable is introduced to establish an acceptable balance between exploration and mining processes. Moreover, the position of beta wolves in the GWO algorithm is used more in deciding to move toward the goal. In this way, inspired by the genetic algorithm, alpha and beta wolves are considered as parents, and 2 children are produced using the crossover, which after checking their fitness is either added to the population and causes the delta wolves to be eliminated or does not affect the process. The RF algorithm is used to calculate and update the fitness value in each iteration of the IGWO-FR method. The proposed technique was assessed through the average number of selected features, average classification accuracy, and best fitness. Additionally, the performance of the proposed algorithm was compared with several popular wrapper evolutionary-based feature selection techniques. Upon experiments and comparisons, it was evident that the suggested IGWO-FR method yielded the most superior results across all the datasets evaluated from the UCI machine learning (ML) repository. Therefore, the utilization of this algorithm for pattern classification was proven to be effective in enhancing classification performance.

In recent years, statistical and data-driven modeling approaches have been increasingly employed to predict element concentrations and to examine relationships among environmental features. In this context, the integration of feature selection techniques with machine learning models enhances model generalization and reduces model complexity by enabling the identification of key elements that are strongly associated with the target feature. This study applies machine learning models to investigate the relationships between Aluminum (Al) and other elements and to predict Al concentration levels in an inland marine ecosystem. Specifically, the study evaluates whether accurate predictions can be achieved using a reduced subset of informative elements rather than the full feature set. The findings demonstrate that machine learning methods, when combined with feature selection, can successfully predict Al concentrations while yielding more interpretable models based on a limited number of significant elements.

Abstract—The COVID-19 pandemic created an unprecedented global health crisis that placed enormous pressure on health- care systems worldwide. Although many infected individuals experience mild symptoms, a considerable proportion of pa- tients develop severe complications including pneumonia, acute respiratory distress syndrome, and multi-organ failure. Early identification of patients who are likely to develop severe illness is therefore critical for improving clinical decision-making and ensuring efficient allocation of healthcare resources. Traditional clinical assessment methods rely on physician experience and laboratory evaluation, which may delay early risk identification. In recent years, artificial intelligence and machine learning techniques have demonstrated strong potential in healthcare analytics and predictive disease modeling. This research presents a predictive modeling framework de- signed to classify COVID-19 patient severity using advanced ma- chine learning techniques. The proposed system analyzes multiple clinical parameters including demographic characteristics, vital signs, laboratory test results, and comorbid conditions. Data preprocessing and feature selection techniques are applied to improve the quality and relevance of the input dataset before model training. Several machine learning algorithms including Decision Tree, Support Vector Machine, Random Forest, and Gradient Boosting are evaluated for severity prediction. The experimental evaluation was conducted using a publicly available COVID-19 clinical dataset containing patient health records and disease outcomes. The performance of the predictive models was assessed using metrics such as accuracy, precision, recall, and F1-score. The results demonstrate that the Gradient Boosting model achieved the highest prediction accuracy of ap- proximately 95%, outperforming other algorithms in identifying severe cases. The proposed framework provides an intelligent decision support tool that can assist healthcare professionals in early risk assessment and patient triage. By enabling accurate prediction of disease severity, the system can contribute to improved patient outcomes and more efficient healthcare resource management during pandemic situations. Keywords: COVID-19 Severity Prediction, Machine Learning in Healthcare, Predictive Analytics, Artificial Intelligence in Medicine, Clinical Data Analysis, Pandemic Risk Prediction. Index Terms—component, formatting, style, styling, insert

The soluble solids content (SSC) of an apple is a critical indicator determining its internal quality and market value. This study proposes a rapid and nondestructive method for apple SSC detection by integrating full visible-near-infrared (Vis-NIR) spectral information. The systematic combination of visible region and NIR region could achieve informational complementarity, with particular focus on the key "red region" spectra bond, which is highly relevant to SSC and physically interpretable. In spectra analysis, classical partial least squares (PLS) regression is adopted to handle spectral multicollinearity, coupled with an innovative genetic algorithm optimized variable importance in projection (GA-VIP) method for feature wavelength selection. VIP ensures the selected wavelengths are physically interpretable, particularly focusing on the red-edge region (∼700nm) highly correlated with SSC, while GA globally optimizes the VIP threshold (converging to 1.492 after 16 generations) rather than using a fixed cutoff. This approach achieves a 90.7% reduction in wavelength count while maintaining predictive accuracy. Our proposed method outperformed the full-spectrum approach and other conventional feature selection techniques, achieving a coefficient of determination (R2) of 0.9523, root mean square error (RMSE) of 0.146%, residual predictive deviation (RPD) of 2.31, ratio of performance to quantile intervals (RPQI) of 2.93, and bias of 0.0181%. More importantly, the method proposed here can be extended to rapid spectral detection in other agricultural products and food safety fields, holding substantial practical application value.

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that is primarily characterized by deficits in social communication and restricted or repetitive behavioral patterns. Although psychologists contribute significantly to the understanding of ASD, offering insights into its cognitive, emotional, and behavioral dimensions through assessments, diagnoses, therapeutic approaches, and family support, the diagnostic process remains complex. This complexity arises from the diverse manifestations of the disorder and the challenges associated with data sharing. In addition, conventional machine learning approaches for ASD detection may struggle with high-dimensional neuroimaging data and may require careful feature engineering. Consequently, this motivated us to enhance ASD diagnosis by incorporating deep learning (DL) techniques for feature extraction alongside a modified exponential-trigonometric optimization (ETO) algorithm as a feature selection (FS) technique. The modified ETO integrates the Arithmetic Optimization Algorithm (AOA) and the Guided Learning Strategy (GLS) to improve diagnostic performance. To evaluate the effectiveness of the proposed model, we utilized resting-state functional MRI (rs-fMRI) data from the Autism Brain Imaging Data Exchange (ABIDE I). Furthermore, the performance of the proposed model was compared with that of established models. The results indicate that the proposed model achieves competitive and, in most cases, superior performance compared with the benchmark methods, demonstrating superior accuracy, sensitivity, and AUC in diagnosing ASD. On average across the three atlas-based feature sets, the proposed model has an accuracy, sensitivity, and AUC of 73%, 78%, and 79%, respectively.

Smartour is an explainable machine learning–based tourist recommendation system designed to provide personalized travel suggestions by leveraging user preferences, travel history, budget, and contextual factors such as location and season. Various machine learning models, including hybrid approaches, were implemented and optimized using data preprocessing and feature selection techniques to enhance performance. Explainability methods were integrated to ensure transparency and help users understand the reasoning behind recommendations, thereby increasing trust in the system. The model achieved an accuracy of 92%, demonstrating its effectiveness in improving user experience and supporting better travel decision-making. Additionally, the system adapts to dynamic user interests and incorporates feedback to continuously refine recommendations, enabling users to discover suitable destinations while reducing planning effort and contributing to more intelligent and user-centric tourism solutions.

This study investigates the effectiveness of various feature extraction methods applied to vibration signals for the automatic detection of production defects in CNC (Computerised Numerical Control) milling machines. A dataset consisting of real-world data collected from CNC machines equipped with accelerometers was used. The objective of the study is to compare three main groups of techniques: time-domain analysis (TDA), frequency-domain analysis (FDA), and time-frequency-domain analysis (TFA). The findings indicate that basic TDA features lack the necessary sensitivity to accurately distinguish between Good Processing (GP) and Bad Processing (BP) states. Frequency-domain methods, such as the Fast Fourier Transform (FFT), median frequency calculation, and the Welch periodogram, provide better insights but still have limitations. The most effective results are obtained with TFA methods, particularly Empirical Mode Decomposition (EMD) and the Hilbert-Huang Transform (HHT), which reveal deeper signal characteristics. Following the feature optimisation studies, it was determined that a combination of four features-FMED, IMF2, IMF5 and WPT26-yielded the optimal performance, with an accuracy of 91.48%. The incorporation of a fifth feature resulted in information saturation within the model and did not improve performance. This study makes a novel contribution to literature by conducting an in-depth investigation into the most effective feature extraction and selection techniques for achieving robust discrimination between GP and BP productions using vibration signals in CNC milling processes. Conclusively, TFA features, supported by advanced signal processing, offer a strong basis for reliable, automated defect detection in CNC milling operations.

Optimizing osteoporosis pre-screening (OOPS) through AI-driven models and validation in the Asian population.

Linear models have long been favored in asset pricing for their interpretability in low-dimensional settings. In the contemporary high-dimensional landscape, however, non-linear machine learning models often deliver superior predictive accuracy, albeit at the expense of interpretability due to their “black-box” nature. This paper proposes and evaluates a novel type of explanatory variable, categorical factors, to reconcile this fundamental trade-off between interpretability and accuracy. These factors are constructed by first classifying a broad set of firm characteristics into distinct groups based on their financial nature and then extracting a single representative factor from each group using tailored feature selection techniques. Using US stock market data, we demonstrate that incorporating these categorical factors significantly enhances the predictive accuracy of linear models. Particularly, when the factors are extracted using non-linear feature selection methods, the performance of linear models can be elevated to a level comparable with or even better than that of advanced machine learning algorithms. As these categorical factors directly linked to distinct aspects of intuitive economic information, this key finding underscores their potential to integrate interpretability with predictive accuracy, thereby advancing both objectives in asset pricing. To further guide practitioners, we provide a framework for analyzing the relative importance of these factors, offering a practical tool for quantitative portfolio management. We also demonstrate how this framework can be used to track the temporal evolution of the factors’ predictive power, which is critical for adapting investment strategies to evolving market regimes.

requirement. Intrusion Detection Systems (IDS) are critical for identifying malicious network behaviors to prevent cyber attacks. Conven- tional methods of using signature-basedintrusion detection have limited capabilities to detect unknown attacks. Machine learning algorithms are prone to various problems such asoverfitting, high false positive rates, and poor performance while handling high-dimensional data.In this context, this paper suggests a hybrid intrusion detec- tion systembased on machine learning and ensemble learning techniques. The system uses various supervised machine learn- ing algorithms to learn from network traffic data. Ensemble techniques are used to enhance the overall detection capabilities of the system by utilizing the advantages of various machine learning algorithms. The suggested system uses acombination of Random Forest as a bagging technique, Gradient Boosting as a boosting technique to enhance sequential learning, and stacking to integrate the results of various base classifiers using a meta classifier. Feature selection techniques are used to remove redundant features to enhance the efficiency of the system. Data balancing techniques such as Synthetic Minority Oversampling Technique (SMOTE) and undersampling are used to handle class imbalanceproblems

Diabetes is a chronic disease characterized by elevated blood glucose levels. Without early detection and proper management, it can lead to serious complications and increase healthcare costs. Its global prevalence is rising, with many cases remaining undiagnosed. In this study, we developed an explainable machine learning model using a two-stage approach for predicting diabetes. Five machine learning (ML) models, including Multi-Layer Perceptron, Support Vector Machine, K-Nearest Neighbor, Extreme Gradient Boosting (XGBoost), and Naïve Bayes, were trained and evaluated using a two-stage approach. In Stage one, a public dataset containing 520 samples was used, and Shapley Additive exPlanations (SHAP) and MLP weights were applied for feature selection. In Stage two, the same models were trained and evaluated using a dataset of 270,943 samples collected from Rwanda. SHAP was further employed to explain the model output. In Stage one, the Multi-Layer Perceptron model achieved the best performance on a public dataset, with an accuracy of 95.19%. Feature selection techniques identified the top 10 influential predictors associated with diabetes risk, including those recommended by diabetes care providers in Rwanda. In Stage two, the XGB model outperformed other models, achieving an accuracy of 97.14%. This study presents a two-stage, explainable machine learning framework for systematic screening for type 2 diabetes. The first stage evaluates risk based on reported symptoms, while the second stage incorporates demographic, anthropometric, and vital sign data for refined risk assessment. Integration of these models into the mUzima mobile application can enhance community health workers' capacity to identify and refer high-risk individuals. By enabling early and accurate detection, the proposed approach has the potential to reduce undiagnosed diabetes and support improved disease management.

In the evolving online education system, Interface Design (ID) plays a crucial role in facilitating the application of Artificial Intelligence (AI)- driven Digital Learning Systems (DLS). While significant research has commonly analyzed educational methods, the accuracy of quantifying the impact of specific Interactive Interface Features (IIFs) on User Experience (UX) remains underdeveloped. This research presents a complete model of the impact of IIF on UX in AI-driven Digital Learning Systems (DLS). This research study employed a controlled experimental design (n = 240) with a between-subjects method to assess five key interface features: Adaptive Feedback Panels (AFP), Gamification Elements (GE), live Conversational Agents (CA), Progress Visualization (PV), and Micro-Assessment Widgets (MAW). This work designed a multi-layer model that precisely manipulated these features while maintaining experimental control. User interactions were analyzed using a Mixed-Methods Approach (MMA), combining Linear Mixed-Effect Modelling (LMEM) with Machine Learning (ML)-based Feature Selection (FS) techniques. Results show that live CA (β = 5.32, p < 0.001) and Adaptive Feedback Mechanisms (AFM) (β = 4.86, p < 0.001) had the most effective positive impact on system usability, while GE most significantly enhanced user engagement (β = 0.42, p < 0.001). The FS method revealed synergistic effects between CA and Adaptive Feedback (AF) (SHAP interaction value = 0.087). ML validated these empirical results, identifying nonlinear relationships and achieving a predictive R² of 0.849 for the composite UX score. This research developed a robust methodological approach for quantifying IIF impacts and provides empirical proof to guide the design of AI-enhanced educational interfaces that optimize learning experiences.

The Air Quality Index (AQI) must be accurately and highly evaluated to limit the effects when exposed to air pollution on individual health and the ecosystem. The traditional techniques are used to predict air quality ineffective in managing the nature of the environmental data being analyzed and minimizing the effects of the most significant features of large numbers of dimensions. The purpose of this work is to address these issues using the creation of a hybrid deep learning algorithm implementing Feature Selection Techniques, Recurrent Neural Networks and a Quantum-Inspired Genetic Algorithm (QIGA). The eXtreme Gradient Boosting (XGBoost) will be applied to the environmental dataset to evaluate importance of features. Additionally, the Principal Component Analysis (PCA) will reduce the dataset's dimensionality using most significant features from the original dataset as inputs for the prediction model. Recurrent Neural Networks (RNNs) are able to detect time-variant patterns of air quality (i.e., the pattern changes over time) based on the use of controlled memory. The use of quantum-based methodologies will allow rapid searches over high-dimensional datasets resulting higher performance than traditional optimization methods. This innovative methodology will lead to improved accuracy, computational efficiency and interpretability of AQI predictions. This method will be the basis for smart environmental monitoring systems used by researchers, policymakers and urban planners to make innovative decisions. Finally, the flexibility of the system makes it possible for users to apply it in any other forecasting-based environmental issues showing the scalability and efficiency.

Future arms control treaties may require new and unconventional approaches to traditional accountancy measures. One possible novel approach is through attribute verification of the isotopic composition of plutonium, with the underlying algorithm utilizing a machine learning model trained on a range of representative gamma spectra. Traditionally, passive gamma-ray spectroscopy for the determination of the 240Pu to 239Pu ratio offers a nonintrusive method of distinguishing weapons-grade plutonium from other forms. However, practical implementation faces challenges due to shielding, unknown configurations, and limited measurement times. To address these challenges, this work examines several feature selection techniques to minimize the information required for accurate weapons-grade plutonium classification. Machine learning models were trained on 5040 synthetic gamma spectra generated with Gamma Detector Response and Analysis Software Detector Response Function, representing a wide range of plutonium masses, ages, isotopic compositions, and shielding scenarios. Random forest classifiers were developed using three feature selection methods: photopeak-based selection, variance-driven channel reduction, and energy truncation. Results demonstrate that models trained on 54 channels in the 632- to 670-keV region achieve strong performance (F1 score +3.27% (95% CI: −4.84%, 11.51%) compared to peak-based benchmark). These findings support the feasibility of machine learning–based attribute verification systems using reduced feature sets while maintaining classification accuracy, with potential integration into treaty-compliant verification frameworks that incorporate information barriers.

Sustainable fishing and aquaculture production require not only data collection but also its proper application. Another factor influencing the study’s direction is dataset size. Regardless of dataset size, the most critical aspect of data use is feature selection. This study focuses on estimating sustainable fishing and aquaculture production by selecting relevant features and appropriate algorithms for both sexes and all individuals in the crayfish population living in the Atikisar Reservoir. For this purpose, the Mean method was used for data preprocessing, Pearson correlation analysis for feature selection, and Multiple Linear Regression (MLR), Support Vector Machine Regression (SVMR), Gradient Enhancement Regression (GBR), Sampled Sum (Bootstrap Combination, Bagging) Regression (BAR), Random Forest Regression (RFR), and k-Nearest Neighbors Regression (k-NNR) algorithms were used for prediction and stock management. Additionally, the data obtained from the population were evaluated within the framework of the classical length-weight relationship. The MLR, GBR, and RFR algorithms showed the highest coefficients of determination (R² = 0.98) for female individuals; the MLR, GBR, BAR, RFR, and k-NNR algorithms had the highest R² values (0.96) for male individuals; and the MLR, GBR, RFR, and k-NNR algorithms achieved the best coefficients of determination (R² = 0.98) for the entire population. Analyzing crayfish populations using the GBR algorithm will enable us to make predictions with the highest accuracy level and to manage populations, after processing the data with the Mean method from data preprocessing methods, and performing feature selection with Pearson correlation analysis. Regression analyses and machine learning algorithms, applied to overall and sexual length and weight of crayfish caught in the Atikhisar Reservoir, revealed positive allometric growth in the TL-TW and meat yield relationship for females, males, and the entire population.