Search Results (1,588)

Predictive analytics has become increasingly important in educational decision-making, supporting at-risk identification and adaptive tutoring. The accurate early prediction of school achievement can enable timely intervention. Using the Math Students dataset, which contains data on students from two Portuguese secondary schools, we model three categorical outcomes derived from the students’ final grade, namely the final grade level (low, medium, high), its qualitative evaluation (fail, satisfactory, good, excellent), and the final pass/fail outcome. After preprocessing, three filter methods—Correlation-Based Feature Subset Selection (CFS), Correlation Attribute Evaluation (CorrEval), and Information Gain (InfoGain)—are applied to reduce the dimensionality of the datasets. Nine classifiers (Naive Bayes, Logistic, MLP, SMO, IBk, Bagging, J48, Random Forest, Random Tree) are evaluated using ten-fold cross-validation in the Waikato Environment for Knowledge Analysis (Weka) platform. Random Forest with InfoGain achieves 90.7% accuracy on the three-band task, while Bagging with InfoGain achieves 92.5% on the binary pass/fail outcome, outperforming benchmarks in prior Educational Data Mining (EDM) studies. Results confirm that prior academic performance indicators (first- and second-period grades) and failure history dominate predictive power and contribute substantially to the success of ensemble models, particularly when paired with feature selection methods that reduce noise and highlight relevant attributes. Full article

Pacemaker generator replacement remains clinically important because battery depletion influences the timing of elective replacement procedures and associated procedural risk. This study investigated whether routinely available pacemaker interrogation-derived telemetry can classify devices with manufacturer-estimated low longevity status, defined as remaining device life below 12 months. A total of 39 Medtronic pacemaker interrogation snapshots were analyzed, including 11 single-chamber, 21 dual-chamber, and 7 triple-chamber CRT-P devices. Four machine learning classifiers, namely, Naïve Bayes, HistGradientBoosting, Logistic Regression, and Random Forest, were evaluated using leave-one-out cross-validation as the primary internal validation strategy, with bootstrap 95% confidence intervals calculated from aggregated out-of-fold predictions. The study is application-based rather than algorithmic, focusing on transparent comparison and interpretation of established classifiers for pacemaker-specific low-longevity classification. Logistic Regression achieved the strongest LOOCV performance, with ROC-AUC 0.941, accuracy 0.872, precision 0.958, recall 0.852, and F1-score 0.902. Random Forest also showed favorable performance, with ROC-AUC 0.855 and F1-score 0.873. Parsimonious baseline analysis showed that months since implantation plus battery voltage achieved ROC-AUC 0.883, accuracy 0.846, and F1-score 0.875, approaching the full Logistic Regression model. SHAP analysis identified months since implantation, battery voltage, and right ventricular capture threshold as the most influential predictors. These findings suggest that routine interrogation variables can classify manufacturer-estimated low-longevity status, but the incremental benefit of full multivariable machine learning over simple predictors was modest. Because the endpoint was based on manufacturer-estimated longevity rather than observed clinical battery failure or generator replacement timing, larger longitudinal and multi-manufacturer validation studies are needed before clinical application. Full article

This paper presents a Python-based v3.12 framework for robotic object sorting in a virtual workcell, combining learning-from-demonstration with a comparative evaluation of classical machine learning classifiers. A user provides a minimal demonstration (e.g., one cube and one cylinder placed into two bins) from which a dynamic type-to-bin rule is inferred. In this study, learning-from-demonstration is implemented at the level of rule acquisition from minimal task examples rather than at the level of trajectory imitation or low-level motion teaching. This rule is used to relabel a larger dataset of pre-generated object positions, enabling training with a selectable number of file-based samples (2–1600) optionally augmented with manual samples. Five classifiers—decision tree, k-nearest neighbors, logistic regression, naive Bayes, and linear SVM—were trained and then used to drive autonomous pick-and-place execution while logging replication time and correctness (correct/incorrect moves and accuracy). Because the task reaches accuracy saturation under a deterministic rule, an additional offline inference benchmark was included to compare prediction throughput using 10,000 probes with repeated timing (median over 50 runs or mean ± standard deviation over 30 runs). To complement this nominal evaluation, the framework also included a perturbation-aware robustness protocol based on controlled positional perturbation, systematic bias, controlled shape corruption, repeated perturbation voting, and stability-aware scoring. This additional layer makes it possible to examine classifier behavior under controlled uncertainty, especially in reduced-data settings, without changing the compact simulator-based nature of the workflow. Results indicate identical sorting accuracy across models, while inference-time differences remain measurable, highlighting deployment-oriented trade-offs and confirming that end-to-end cycle time is dominated by robot motion rather than model computation. Full article

Automated short-answer grading (ASAG) in educational contexts faces a fundamental trade-off between predictive performance, interpretability, and methodological transparency, particularly under data-constrained educational settings. While recent approaches rely on deep learning architectures, these models require large annotated datasets and offer limited transparency, restricting their applicability in authentic classroom environments. This study proposes a fully specified and interpretable machine learning pipeline for ASAG across multiple educational concepts. The approach is based on a shared TF–IDF representation and evaluates three linear classifiers—Logistic Regression, Multinomial Naïve Bayes, and Linear Support Vector Machines—under a stratified cross-validation framework adapted to small datasets. Model performance is assessed using accuracy, precision, recall, and F1-score. Statistical comparisons using the Wilcoxon signed-rank test indicate exploratory evidence of statistically significant differences between classifiers, although the observed differences remain small in practical magnitude. Additionally, the methodology incorporates token-level analysis to identify discriminative lexical patterns and examine consensus across classifiers. To enhance interpretability, tokens are presented using a bilingual Spanish/English representation while preserving the original feature space. The results across ten concept-specific datasets show consistent performance across models (accuracy ≈ 0.82–0.88) and reveal stable lexical patterns consistently associated with model predictions of correctness. The findings highlight that lightweight, interpretable models can provide consistent and reliable performance under resource-constrained educational conditions. The proposed framework contributes a stability-oriented and interpretable evaluation paradigm for ASAG, offering a practical alternative to data-intensive approaches in educational assessment. It is intended as a methodological reference protocol rather than a performance benchmark. The findings should be interpreted as evidence of within-context consistency instead of broad external generalization. Full article

The geographical origin of aniseed (Pimpinella anisum L.) represents a key quality determinant, as it directly influences the chemical composition and commercial value of its essential oil. Agronomic traits of aniseed (plant height, umbel diameter, number of umbels per plant), productivity-related traits (number of seeds, thousand-seed weight, yield per plant, plant biomass, harvest index, yield per hectare, essential oil content and yield), and physiological traits (germination energy and total germination) exhibit variations depending on geographical origin. The study proposes an integrated framework for accurate classification by combining agronomic, productivity, and physiological data with GC-MS profiles and advanced machine learning (ML) techniques. A total of 144 samples were analyzed, based on a factorial design including three locations, six fertilizer treatments, two years, and four replications. trans-Anethole was the dominant compound in all samples (89.508–101.441%). Several classification models, including artificial neural networks, random forests, MARSplines, boosted trees, interactive trees, naïve Bayes, and support vector machines, were evaluated to discriminate samples by geographical origin using agro-meteorological and GC-MS data. The results indicate that AI and ML approaches effectively captured complex non-linear relationships. Overall, the multi-model framework highlights the strong potential of machine learning for agro-food authentication, supporting improved traceability, site-specific decision-making, and quality control. Full article

Flooding is a complex, unpredictable disaster that occurs frequently and can have devastating impacts. Over the past two decades, the advent of machine learning (ML) methods has led to a surge in studies focused on flood prediction, emphasizing high-performance algorithms and fast processing times. The present study aims to investigate the challenges of generalization in flood prediction models using machine learning techniques. A dataset of 18,810 samples was compiled from 40 river basins covering the period 1959–2020. Nine machine learning algorithms were applied to the analysis: Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, AdaBoost, Gradient Boosting, Extra Trees, and Gaussian Naive Bayes. Four distinct validation methods were employed to assess the performance of the models, and the results were thoroughly analyzed. The Gradient Boosting model demonstrated exceptional validation performance indicating its robustness across diverse datasets. High accuracy was also observed in the Decision Tree, Random Forest, Extra Trees, and AdaBoost models. However, for datasets with fewer than 200 samples, these four models experienced a decline in performance. Elevation was identified as the most important factor influencing flooding in 36 basins. NDVI was the dominant factor in 3 basins, while rainfall was the main driver in only 1 basin. The results highlight the contributions and shortcomings of machine learning methods in sustainable flood disaster management systems. Full article

Background/Objectives: Early transient endocrine dysfunction after simultaneous pancreas-kidney transplantation (SPK) frequently triggers urgent investigations to exclude thrombosis, pancreatitis, or rejection, yet many recipients recover during the index admission. We tested whether pre-transplant day zero (D0) serum Fourier-transform infrared spectroscopy (FTIRS) captures a biochemical fingerprint associated with a Start&Stop trajectory (initial insulin independence followed by transient dysfunction with recovery). Methods: In a single-center retrospective case-control study nested within 104 consecutive SPK recipients with available D0 serum, 12 Start&Stop cases were matched 1:1 to 12 No-Stop controls. Serum FTIR spectra went through structured quality control and standardized preprocessing. A Naïve Bayes classifier with Fast Correlation-Based Filter (FCBF) feature selection was evaluated using leave-one-out cross-validation (LOOCV) and label-permutation analysis. Results: Under LOOCV, the primary FTIRS model (Savitzky-Golay second derivative; 600–900 and 2800–3400 cm⁻¹) achieved excellent discrimination (ROC-AUC 1.00) with accuracy 0.958 and F1 score 0.958. Discrimination collapsed under label permutation (ROC-AUC 0.461), supporting a non-random label-spectrum association. Discriminant information mapped mainly to carbohydrate/glycoprotein-associated bands (~946–1161 cm⁻¹), protein structural contributions near the amide III region (~1300 cm⁻¹), and lipid/protein stretching modes (~2865–3163 cm⁻¹), consistent with a multicomponent systemic biochemical state. Conclusions: In this exploratory matched case-control cohort, pre-transplant D0 serum FTIRS signatures were associated with the subsequent Start&Stop phenotype after SPK. These findings should be interpreted as recipient-side exploratory risk-stratification signals rather than clinically actionable decision tools. Larger multicenter validation in unselected cohorts, with standardized endpoint adjudication, preanalytical control, fully nested model development and inter-instrument harmonization, is required before clinical implementation or population-level risk calibration. Full article

Zero Trust architectures require immediate identification of IoT devices before granting network access; however, most existing classification methods rely on extended traffic observation windows or computationally intensive deep learning models. This study proposes a lightweight multi-label IoT device classification framework based solely on early-stage DHCP and DNS metadata captured during device boot-up. Traditional supervised classifiers, including Naïve Bayes, Decision Tree, Random Forest, and Multi-Layer Perceptron, are adapted to support probabilistic multi-label prediction and integrated unknown device detection through confidence-based thresholding. The approach enables devices with identical or overlapping behavioral fingerprints to be grouped for policy enforcement while preserving detection sensitivity for unseen devices under open-set conditions. Experimental evaluation on 40 IoT devices representing 31 device types demonstrates that Random Forest achieves the most reliable balance between classification accuracy and unknown detection robustness, while maintaining low computational overhead suitable for constrained gateways. The results show that early metadata alone is sufficient for real-time Zero Trust enforcement and least-privilege policy activation. The proposed unified framework reduces architectural complexity by combining classification and unknown detection into a single model, making it practical for scalable IoT deployments. Full article

The exponential growth of genomic and proteomic data has made computational protein–protein interaction (PPI) prediction indispensable, driving the need for a comprehensive and method-aware evaluation of supervised learning approaches. PPIs are fundamental to understanding cellular processes and disease mechanisms, yet experimental identification remains slow, costly, and difficult to scale. This survey systematically investigates ten supervised learning models—Extreme Learning Machine (ELM), Convolutional Neural Networks (CNNs), Graph Neural Networks (GNNs), Deep Neural Networks (DNNs), Naïve Bayes, Probabilistic Decision Tree, Support Vector Machine (SVM), Least Squares SVM (LS-SVM), K-Nearest Neighbor (KNN), and Weighted K-Nearest Neighbor (WKNN)—through a tri-layered framework that integrates Comparative Quantitative Analysis, Comparative Observational Analysis, and Experimental Evaluations. Beyond conventional accuracy summaries, this work provides critical commentary tied to real-world use, analyzing where techniques succeed or fail in practice—for instance, when instance-based methods bottleneck during inference, when kernel choices influence SVM variance, or when deep architectures trade accuracy for computational cost. The survey also offers concrete deployment guidance, such as calibration insights for WKNN versus KNN under varying feature noise or dataset curation quality, delivering operational perspectives that typical surveys omit. Comparative Quantitative Analysis consolidates metrics such as accuracy, F1-score, and computational time from the existing literature, while Comparative Observational Analysis evaluates interpretability, scalability, dataset suitability, and efficiency. Complementing these, Experimental Evaluations conducted by the authors empirically validate model performance on benchmark datasets. Together, these layers provide a unified and evidence-backed perspective on algorithmic strengths, weaknesses, and practical applicability. Findings show that GNNs and DNNs achieve the highest predictive accuracy due to their ability to capture structural and topological relationships, whereas ELM and Naïve Bayes offer superior efficiency. SVM and LS-SVM maintain robust stability under noisy conditions, and CNNs are well-suited for sequence-based prediction tasks. By combining empirical validation, critical insights, and deployment-focused recommendations, this survey delivers decision-grade guidance that bridges theoretical understanding with real-world implementation, thus clarifying the trade-offs among accuracy, efficiency, and scalability in PPI detection research. Full article

Fetal health assessment is a crucial aspect of prenatal care, aimed at the early detection of potential complications to ensure optimal outcomes for both mother and child. Traditional methods, such as the visual analysis of cardiotocography (CTG) data by healthcare professionals, are valuable but often subjective and time-consuming. This work investigates the application of machine learning techniques, with a focus on ensemble learning, to enhance the accuracy and efficiency of fetal health classification based on CTG data. Genetic Algorithm (GA) is employed for optimal feature selection, identifying the most discriminative subset of CTG attributes to improve model performance and reduce computational complexity. We employ a combination of advanced machine learning models, including AdaBoost, Gaussian Naïve Bayes, Decision Tree, k-nearest neighbors (KNN), and Logistic Regression. The top two models were selected based on comprehensive performance metrics using the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method. These models were then integrated through ensemble learning approaches, such as stacking, Particle Swarm Optimization (PSO) weighted averaging, and soft voting, to improve prediction reliability. Our proposed stacking ensemble model achieves a remarkable accuracy of 97.9%, demonstrating its potential as a robust, data-driven tool for fetal health monitoring and the early identification of at-risk pregnancies. The results indicate that machine learning can effectively complement traditional fetal health assessment methods by providing an objective framework to support clinical decision-making. Full article

Glamping tourism has expanded rapidly as travelers increasingly seek nature-based experiences combined with comfort and privacy, particularly in the post-COVID-19 period. Online reviews provide a valuable source of insight into how guests perceive such experiential accommodation, yet large-scale, data-driven analyses of glamping sentiment remain limited. This study applies machine-learning techniques to classify customer sentiment expressed in online reviews of glamping sites in South Korea. A total of 3233 reviews were collected from ten leading glamping locations on Naver Map, cleaned, and translated from Korean to English. Sentiment labels (negative, neutral, and positive) were generated using VADER (Valence Aware Dictionary and sEntiment Reasoner), a lexicon-based sentiment scoring tool validated for short informal texts and the labeled corpus was subsequently used to train and evaluate six supervised classifiers. Six supervised classifiers—Naïve Bayes, k-Nearest Neighbors, Random Forest, Logistic Regression, Gradient Boosting, and Support Vector Machine (SVM)—were trained and evaluated through stratified ten-fold cross-validation using accuracy, AUC, F1-score, and Matthews Correlation Coefficient (MCC). Results indicate that SVM achieved the strongest overall discriminatory performance, particularly in identifying minority sentiment classes under substantial class imbalance. These findings suggest that automated sentiment classification holds practical potential for supporting evidence-based service monitoring and reputation management in glamping tourism, although further validation in operational settings is needed before deployment can be recommended. Full article

Background: Dengue fever remains a major public health challenge in Mexico, exhibiting pronounced seasonal behavior and substantial geographic heterogeneity. Using recent epidemiological surveillance data may improve predictive performance and better reflect the current epidemiological context. Objective: The aim of this study was to develop and compare temporal machine learning models for the binary classification of confirmed and negative dengue cases in Mexico using 2025 national surveillance data. Methods: A total of 68,222 suspected dengue cases reported in 2025 were analyzed. The outcome variable was CASE_STATUS, encoded as 0 for negative cases and 1 for confirmed cases. The dataset was divided chronologically into training (January–September), validation (October), and testing (November–December) subsets. Nine machine learning algorithms were evaluated: Random Forest, Bayesian Network, XGBoost, CatBoost, Naïve Bayes, Logistic Regression, Multilayer Perceptron, Support Vector Machine, and LightGBM. Preprocessing included scaling, encoding, age discretization for Bayesian Network, class imbalance treatment, and model-specific feature-importance analyses. Performance was assessed using accuracy, Precision, Recall, F1-score, ROC-AUC, and PR-AUC. Results: Random Forest achieved the best overall performance, with the highest test F1-score (0.7254) and PR-AUC (0.7300) at an optimized threshold of 0.397, together with a high Recall (0.8938). Bayesian Network achieved the highest test accuracy (0.7023) and ROC-AUC (0.7756), although its overall operational balance was less favorable considering class imbalance. Geographic and institutional variables were the most influential predictors across models, whereas comorbidities generally contributed less. Conclusions: Temporal machine learning models are useful for dengue case classification in Mexico, and Random Forest was the most robust approach, balancing sensitivity and overall predictive performance. From an operational perspective, this finding is especially relevant in dengue surveillance, where failure to identify true confirmed cases may have important public health consequences. Full article

Machine learning helps intrusion detection systems learn new assaults quickly. These systems train on a dataset with several threats and may identify odd behavior. This research detects intrusion using Random Forest, KNN, and Gaussian Naive Bayes. We run the model on a comprehensive dataset. Dynamics Feature Selector (DFS) improves performance. This technique eliminates unnecessary inputs and improves predictions using statistical analysis and feature significance. DFS effectiveness is tested using the NSL-KDD dataset. The recommended hybrid approach, Gaussian NB, Random Forest, and KNN are compared in meta-learning. Getting excellent accuracy with fewer characteristics is the aim. In order to demonstrate how the model may function in actual cybersecurity scenarios, the final test makes use of common performance metrics such as accuracy, precision, recall, and F1-score. The proposed method outperforms previously reported results with around 96.09% accuracy, 93.21% precision, 92.53% recall, 92.79% F1-score, and 93.65% average performance. Full article

Background/Objectives: A blunted growth hormone (GH) response in stimulation tests (GHSTs) in obese patients is well documented, with less evidence for insulin-like growth factor-1 (IGF-1) concentrations. The aim of this study was to assess the relationships between nutritional status, GH peak in GHST, and IGF-1 concentrations, and to develop machine learning prediction models of GH deficiency (GHD) in children with short stature. Methods: A case–control study included 1592 children with short stature, whose height, weight, body mass index (BMI), GH peak in two GHSTs, IGF-1 concentration and bone age (BA) were assessed. The cut-off of GH peak in two GHSTs between GHD and idiopathic short stature (ISS) was 10.0 µg/L; additionally, a lower cut-off of 7.0 µg/L was used in repeated analysis. Univariate statistical analyses and classification models were used to identify variables related to the normal and subnormal results of GHST. Results: Depending on the cut-off of GH peak (10.0 vs. 7.0 µg/L), GHD was diagnosed in 604 vs. 279 patients (37.9% vs. 17.5%). Children with GHD had significantly lower (p < 0.001) BMI SDS and IGF-1 SDS than ones with ISS for both cut-offs of GH peak. Overnutrition was associated with the lowest GH peak but the highest IGF-1 SDS; the opposite results were observed in undernutrition. A decision tree predicted GHD in 156 patients, in 149 based on BMI SDS > 0.91. A Naïve Bayes classifier predicted GHD in 118 cases, with BMI SDS and IGF-1 SDS being the only significant variables. The best multilayer perceptron (MLP) neural network predicted GHD in 310 patients, while a logistic regression model did so in 269 patients. Conclusions: Interpretation of GHST should include the patient’s nutritional status in order to avoid overdiagnosis of GHD in overweight and obese children. Full article

The growing complexity of unwanted messages, especially SMS spam, presents a serious challenge to the security of digital communication and user experience. While conventional spam detection models are useful on clean datasets, they are vulnerable to targeted attacks that aim to evade detection. This study is motivated by the urgent need to evaluate the resilience of machine learning models against evolving threats in real-world applications. We specifically investigate the robustness and interpretability of a Multinomial Naive Bayes (MNB) model, representative of traditional machine learning, and a Tiny Text convolutional neural network (Tiny Text CNN), representative of deep learning models, for SMS spam detection. Using the UCI dataset under simulated adversarial text attacks, both models were tested against filler-word insertion and character-level perturbation attacks. Results show that while the Tiny Text CNN maintained higher overall robustness (accuracy: 0.9821 clean vs. 0.9758 under character attacks), both models experienced notable degradation in recall, with MNB being more susceptible to filler-word attacks. Interpretability analyses using LIME and gradient-based saliency maps indicated that adversarial perturbations alter feature importance, diminishing the influence of spam-indicative tokens. The findings underscore the trade-offs between model complexity and adversarial resilience, offering insights for developing more secure and interpretable spam detection systems. Full article