Search Results (4,682)

Virtual Learning Environments (VLEs) have become increasingly popular in education, particularly with the rise of remote learning during the COVID-19 pandemic. Assessing student performance in VLEs is challenging, and the accurate prediction of final results is of great interest to educational institutions. Machine learning classification models have been shown to be effective in predicting student performance, but the accuracy of these models depends on the dataset’s size, diversity, quality, and feature type. Class imbalance is a common issue in educational datasets, but there is a lack of research on addressing this problem in predicting student performance. In this paper, we present an experimental design that addresses class imbalance in predicting student performance by using the Synthetic Minority Over-sampling Technique (SMOTE) and Generative Adversarial Network (GAN) technique. We compared the classification performance of seven machine learning models (i.e., Multi-Layer Perceptron (MLP), Decision Trees (DT), Random Forests (RF), Extreme Gradient Boosting (XGBoost), Categorical Boosting (CATBoost), K-Nearest Neighbors (KNN), and Support Vector Classifier (SVC)) using different dataset combinations, and our results show that SMOTE techniques can improve model performance, and GAN models can generate useful simulated data for classification tasks. Among the SMOTE resampling methods, SMOTE NN produced the strongest performance for the RF model, achieving a Region of Convergence (ROC) Area Under the Curve (AUC) of 0.96 and a Type II error rate of 8%. For the generative data experiments, the XGBoost model demonstrated the best performance when trained on the GAN-generated dataset balanced using SMOTE NN, attaining a ROC AUC of 0.97 and a reduced Type II error rate of 3%. These results indicate that the combined use of class balancing techniques and generative synthetic data augmentation can enhance student outcome prediction performance. Full article

Background: The endoscopic approach to the spine has a steep learning curve, primarily due to challenges in learning how to access Kambin’s triangle, leading to significant radiation exposure. Real-time live instrument tracking using two-dimensional fluoroscopic navigation has been shown to significantly reduce the time and radiation exposure needed to perform a trans-Kambin approach; however, the impact on the learning curve for those in training has not been studied. Methods: Physicians in training (PITs) in a single program were evaluated while accessing Kambin’s triangle using a gel sawbone model. The PITs were randomized in terms of spinal level, the technology used first, and the approach side. Time to access Kambin’s triangle and radiation were recorded for each cohort. A linear regression model was used to assess associations between procedure time and radiation exposure and endoscopic experience and PGY level. Results: Seven PITs were studied with a range of prior experience, consisting of three PITs with high exposure (HE) (≥30 cases) and four with low exposure (LE) (0–20 cases, averaging 7.5). Unassisted by instrument tracking, the SE group was 63% faster and was exposed to 63% less radiation. Further, there was a moderate correlation with experience and time (R = −0.52) and radiation (R = −0.60). Using instrument tracking allowed all the PITs to be 40% faster, taking 6.6 min (range 2.9–12.4) without instrument tracking and 4.0 min (1.9–6.3) with it. Radiation also decreased by 91% (2.52 vs. 0.23 mGy with instrument tracking). Conclusions: Experience significantly enhanced the accessibility of Kambin’s triangle, reducing time and radiation exposure by about two-thirds. Regardless of experience, instrument tracking reduced radiation exposure by 90%. Additionally, familiarizing PITs with instrument location can eliminate 70% of the learning curve for those inexperienced with accessing Kambin’s triangle. Full article

The present study investigates simple cubic lattice structures fabricated through an FDM-based three-dimensional (3D) printing method using wood–polylactic acid (wood–PLA) bio-composite filament and develops a data-driven framework to predict their mechanical response. The design of experiments (DOE) was developed using a response surface methodology (RSM) based on a central composite design (CCD) that was implemented in Design-Expert software (Version 13). During fabrication, four different manufacturing parameters—the layer height, the printing speed, the nozzle temperature, and the infill density—were considered. The compressive strength and compressive modulus were evaluated experimentally, and the corresponding stress–strain responses were examined. The results reveal that the layer height is the most influential parameter, where lower layer heights (0.06–0.1 mm) significantly improve both the compressive strength and the modulus due to enhanced interlayer bonding and reduced void formation. The printing speed and the nozzle temperature also play critical roles, where lower printing speeds (≈40 mm/s) and moderate nozzle temperatures (≈195–205 °C) promote more uniform material deposition and improved interlayer bonding, while higher speeds (≥60 mm/s) and excessive temperatures (≈225 °C) lead to reduced bonding quality and a deterioration in mechanical performance. In contrast, the infill density exhibited a non-monotonic influence, where intermediate levels (around 70%) provided an improved performance under combinations of the low layer height (≈0.1 mm), the low printing speed (≈40 mm/s), and the moderate nozzle temperature (≈195–215 °C), suggesting an interaction-driven effect rather than a purely density-dependent trend. To complement the experimental findings, a machine learning model based on eXtreme Gradient Boosting (XGBoost) was developed using 12,000 data points that were derived from stress–strain curves. The model successfully predicted continuous mechanical responses with errors in the range of 2–8% for unseen specimens, suggesting its capability to capture the relationship between printing parameters and mechanical behavior within the studied design space. Overall, the study highlights that the mechanical properties of wood–PLA lattice structures can be effectively tailored by choosing an appropriate printing parameter control and demonstrates the feasibility of using machine learning to estimate mechanical performance without additional physical testing within the defined parameter domain. Full article

Unmanned Aerial Vehicle (UAV)-assisted hierarchical federated learning (HFL) has emerged as a promising architecture for Internet of Things (IoT)-based smart agriculture, which enables scalable model training over large and sparse farmlands. In this setting, UAVs act as mobile edge servers, aggregating local updates from distributed agricultural IoT devices and relaying them to the cloud server. While HFL improves scalability and reduces communication overhead, it still faces critical security threats due to its reliance on public wireless channels and the vulnerability of model aggregation to malicious updates. In this paper, we propose a secure authentication scheme that integrates anomaly detection with elliptic curve cryptography (ECC)-based mutual authentication to protect both the communication and training phases. In the proposed scheme, UAVs authenticate participating clients before receiving their local models, then perform anomaly detection to identify and exclude malicious participants. If a client is found to be malicious, its identity credentials are revoked and broadcast by the cloud server to prevent future participation. The security of the proposed scheme is formally verified using Burrows–Abadi–Needham (BAN) logic, the Real-or-Random (RoR) model, and the Automated Validation of Internet Security Protocols and Applications (AVISPA) tool, along with informal security analysis. The performance evaluation includes comparisons of security features, computation cost, and communication cost with other related schemes, and an experimental assessment of anomaly detection performance. The results demonstrate that our scheme provides strong security guarantees, low overhead, and effective malicious client detection, making it well suited for UAV-assisted HFL in smart agriculture. Full article

Log anomaly detection in modern distributed systems is challenging. Anomalous behaviors are rare. Manual labeling is expensive. Session boundaries are often set by fixed heuristics before model training. This fixed-boundary assumption is problematic because segmentation errors propagate into representation learning and cannot be corrected during optimization. To address this, this paper proposes BASN (Boundary-Aware Sessionization Network), a boundary-aware contrastive learning framework that jointly learns session boundaries and anomaly representations using a differentiable soft-reset mechanism. BASN does not treat sessionization as a separate step. Instead, it predicts boundary probabilities from event semantics and temporal gaps, then modulates end-to-end session-state updates. The session representations are optimized with self-supervised contrastive learning, enabling effective zero-shot anomaly detection and few-shot adaptation. Experiments on four benchmark datasets (BGL, HDFS, OpenStack, SSH) show strong zero-shot performance (area under the receiver operating characteristic curve, AUROC 0.935–0.975) and boundary alignment with expert-validated proxy segmentation (boundary F1 0.825–0.877). Comparative gains over baselines are reported in the article after bibliography correction, baseline verification, and expanded statistical analysis. BASN is also computationally efficient, requiring less than 10 ms per session on a Graphics Processing Unit (GPU) and less than 45 ms on a Central Processing Unit (CPU). This is compatible with real-time inference needs in the evaluated settings. However, cross-system transfer AUROC (0.735–0.812) remains below in-domain performance. Domain-specific adaptation is still needed for deployment in environments that differ greatly from the training domain. Full article

The rapid expansion of global commerce has escalated the complexity of money laundering schemes, making the detection of illicit transfers an urgent but highly challenging research problem. In operational anti-money laundering (AML) systems, the extreme rarity of illicit transactions often overwhelms compliance teams with false positives, leading to severe “alert fatigue.” To address this critical bottleneck, this paper introduces an enhanced, probability-driven risk-prioritization framework utilizing an XGBoost classifier integrated with Bayesian Optimization (BO-XGBoost). By optimizing directly for the Area Under the Precision–Recall Curve (PR-AUC), the model is specifically tailored to rank high-risk anomalies under severe class imbalance. We validate the proposed approach on a rigorously resampled transaction dataset simulating a realistic 5% laundering rate. The BO-XGBoost model demonstrates exceptional prioritization capability, achieving an ROC-AUC of 0.9686 and a PR-AUC of 0.7253. Most notably, it attains a near-perfect Precision@1%, meaning the top 1% of flagged transactions are 100% true illicit activities, entirely eliminating false positives at the highest priority tier. Comparative and SHAP-based interpretability analyses confirm that BO-XGBoost easily outperforms sequence-heavy deep learning baselines. Crucially, it matches computationally expensive stacking ensembles in peak predictive precision while significantly surpassing them in operational efficiency, indicating its immense promise for resource-optimized, real-world compliance screening. Full article

The nonlinearity and the extreme variability of seismic signals makes the detection of earthquakes difficult. Although the conventional deep-learning models can be used to extract useful features, they cannot be used in early-warning systems due to their non-interpretability. In this study, a Bidirectional Long Short-Memory network with an attention system (Bi-LSTM-Attn) is proposed to detect seismic events using the ConvNetQuake dataset. To improve the quality of data, the entire preprocessing pipeline, such as signal filtering, segmentation, normalization, and noise reduction is employed. The model was optimized using hyperparameter tuning of sequence length, learning rates, and attention weighting to achieve the best number of true-positive detections and a minimum number of false alarms. The accuracy, precision and recall, F1-score, MSE, and ROC curves were used to assess the performance and the attention weight visualization allowed interpreting the model. It is proven through experiments that the Bi-LSTM-Attn model provides more credible and comprehensible forecasting in relation to baseline LSTM and GRU models. Making the high-accuracy classification and the transparent decision behavior, the approach will increase the level of trust to the automated seismic surveillance, as well as help to build the reliable global networks of earthquake early-warnings. Full article

Background/Objective: Chest radiography (CXR) is routinely used in the evaluation of respiratory disease; however, differentiating COVID-19 from other viral pneumonias on CXR remains challenging due to substantial radiographic overlap. In this study, a deep learning-based CXR classification model using a ResNet-50 architecture was developed to categorize images as normal, COVID-19, or non-COVID viral pneumonia, with emphasis on bias mitigation and external validation. Methods: Model training and internal validation were performed using harmonized publicly available datasets with patient-level stratified five-fold cross-validation, while generalizability was evaluated using an independent real-world institutional dataset from Adan Hospital, Kuwait, which was excluded from all training, validation, and hyperparameter tuning stages. Results: On the public validation dataset (n = 847), the model achieved an overall accuracy of 96.8% with balanced class-wise performance, whereas performance on the independent institutional dataset (n = 320) decreased to 93.7%, consistent with expected domain shift. Calibration analyses demonstrated well-aligned probabilistic estimates on validation data and acceptable calibration on institutional data. Negative predictive values remained high for normal and COVID-19 classes across datasets. Exploratory decision curve analysis demonstrated net benefit patterns for COVID-19 predictions under hypothetical threshold assumptions. Conclusions: These findings indicate that, when developed with explicit bias-mitigation strategies and evaluated using independent institutional data, deep learning-based CXR analysis may provide supportive, non-diagnostic decision signals for radiology triage workflows; however, prospective multicenter validation is required prior to clinical adoption. Full article

The accurate prediction of motor progression in Parkinson’s disease (PD) remains a major clinical challenge that limits personalized treatment planning and efficient clinical trial design. In this study, we developed and validated a machine-learning framework integrating a targeted panel of plasma proteins measured by Olink proximity extension assays with clinical variables to stratify patients according to their progression risk. We analyzed baseline plasma samples from 211 early-stage PD patients enrolled in the Parkinson’s Progression Markers Initiative (PPMI) cohort using four targeted Olink panels, from which 28 circulating proteins were retained after quality-control filtering. Patients were classified as rapid or slow progressors based on their annualized change in MDS-UPDRS Part III scores. Among the algorithms tested, Random Forest achieved the highest discriminative performance with an area under the receiver operating characteristic curve (AUC) of 0.751 (95% CI: 0.684–0.811), which exceeded that of clinical predictors alone (AUC 0.666). The integration of targeted proteomic and clinical features further improved model performance (AUC 0.773; p = 0.009). Nested cross-validation confirmed minimal optimistic bias (AUC 0.743). To enhance clinical interpretability, we applied SHapley Additive exPlanations (SHAP) analysis, which identified interleukin-6 (IL-6), brain-derived neurotrophic factor (BDNF), and vascular endothelial growth factor A (VEGF-A) as the most influential predictors. SHAP feature rankings were highly stable across cross-validation folds (mean Spearman ρ = 0.91). The robustness of these findings was confirmed through sensitivity analyses using extreme quartile comparisons (AUC 0.823), treatment-naïve subgroup analysis (AUC 0.738), and a clinically anchored outcome definition based on the minimal clinically important difference (AUC 0.739). A decision curve analysis demonstrated a net clinical benefit across threshold probabilities of 0.25–0.70. Our results establish targeted plasma protein profiling combined with interpretable machine learning as a promising tool for PD motor progression risk stratification, with potential applications in individualized patient counseling regarding motor prognosis and the selection of candidates for disease-modifying trials. Full article

Knowledge Tracing (KT) is a fundamental task in intelligent education systems, designed to track students’ evolving knowledge states and predict their future performance. While Deep Learning-based Knowledge Tracing (DLKT) models have advanced the field, they often face significant limitations in jointly capturing short-term performance fluctuations and long-term knowledge retention, which restricts their predictive precision in complex learning trajectories. This paper proposes the Extended Deep Knowledge Tracing (xDKT) model, which integrates the Extended Long Short-Term Memory (xLSTM) architecture to enhance multi-scale temporal learning representations. Specifically, through rigorous ablation studies over extended learning sequences (up to 1000 steps), our analysis indicates that the exponential gating and advanced scalar memory of sLSTM units are the primary drivers of performance. This architecture effectively captures both short-term performance shifts and long-term knowledge retention without the vanishing gradient degradation inherent to standard LSTMs. We evaluate xDKT across six diverse benchmark datasets, including Synthetic, Algebra2005–2006, Statics2011, and the ASSISTments series, covering over 22,000 learners. Experimental results show that xDKT yields improved Area Under the ROC Curve (AUC) scores on Statics2011 (0.8562) and ASSISTments2009 (0.8318) compared to baseline models such as DKT, DKVMN, and AKT. Finally, through extensive validation, these findings suggest that xDKT architecture provides a robust and promising framework for accurate and adaptive learning environments. Full article

Background: Postoperative complications following colorectal surgery remain a persistent clinical challenge. Traditional risk stratification has focused on patient characteristics, while conventional modeling approaches treat procedural factors such as operative duration and surgical approach as independent predictors, potentially obscuring interaction effects. Methods: This study developed a machine learning model stratifying 7908 colorectal surgery patients into four distinct procedural risk categories based on combined surgical approach and operative duration (laparoscopic-short, laparoscopic-long, open-short, open-long), rather than treating these factors as separate variables. A gradient boosting ensemble classifier with RUSBoost resampling was trained on predictor variables including patient demographics, comorbidities, and intraoperative factors. Results: Feature importance analysis revealed that the open-long category emerged as the single most important predictor, substantially exceeding all other variables. Weight loss, body mass index, patient age, and electrolyte abnormalities ranked as the next most important predictors. Stratified complication rates demonstrated a critical interaction: prolonged duration more than doubled complication risk in open procedures (short-duration: 9.99%, long-duration: 20.46%), whereas laparoscopic procedures showed only a modest increase from short-duration (10.45%) to long-duration (14.08%) cases. Logistic regression benchmark analysis confirmed the duration-approach interaction (OR = 1.53, 95% CI: 0.97–2.39), achieving comparable discrimination (c-statistic 0.678 vs. 0.665 for the ensemble model). Decision curve analysis demonstrated logistic regression provided superior clinical utility across most threshold probabilities. Conclusions: The dual analytical framework (i.e., statistical inference for quantifying associations and machine learning for predictive feature ranking) offers complementary insights for clinical application. These findings demonstrate that stratified feature engineering can elucidate complex risk phenotypes that may be obscured when procedural factors are analyzed independently. Full article

Background: Predicting genomic alterations from routine hematoxylin and eosin (H&E) whole-slide images (WSIs) may help triage molecular testing. Methods: We retrospectively enrolled 437 patients at Osaka Metropolitan University Hospital across 26 cancers, matched with clinical gene-panel data. We curated 1023 binary endpoints across SNV, CNV, and SV categories. We extracted slide embeddings from five pathology foundation models (Prism, GigaPath, Feather, Chief, and Titan) using a unified feature extraction pipeline and benchmarked them using a lightweight downstream Multi-Layer Perceptron (MLP) classifier. Using the best-performing patch feature system, we trained a multi-instance learning model to assess incremental benefit. Results: Titan achieved the highest and most stable transfer performance, with a median endpoint-wise Area Under the Receiver Operating Characteristic curve (AUROC) of 0.77 in the slide benchmarking; at the patch-level, prediction of APC_SNV reached an AUROC of 0.916, and prediction of KRAS_SNV reached an AUROC of 0.811 on the held-out test set. Conclusions: In a heterogeneous clinical gene-panel setting, pathology foundation models can provide strong baseline genomic-prediction signals without additional fine-tuning. We propose a practical, deployment-oriented two-stage workflow: rapid slide-embedding screening to prioritize robust representations and candidate endpoints, followed by patch-level training for high-value tasks where additional performance gains and interpretable regions are clinically worthwhile. Full article

Background/Objectives: This study aimed to automatically classify Zirconia-based fixed restorations and porcelain-fused-to-metal (PFM) restorations on panoramic radiographs using an artificial intelligence-based model. Unlike previous studies that mainly focused on classifying types of restorations (e.g., crowns, fillings, implants), this research concentrated on material-based differentiation, aiming to provide a more specific contribution to clinical decision support systems. Method: Panoramic radiographs obtained from the archive of Süleyman Demirel University Faculty of Dentistry were included in this study. Radiographs with poor image quality or insufficient visibility of the restoration area were excluded. A total of 593 cropped region-of-interest (ROI) images, labeled by expert prosthodontists using ImageJ software (version 1.54r; National Institutes of Health, Bethesda, MD, USA), were included in the analysis. In order to reduce class imbalance, data augmentation was applied only for images in the Zirconia-based fixed restorations class. By using various image processing techniques such as rotation, reflection and brightness change, the number of samples in the zirconia-based restorations class was increased and thus a balanced dataset was obtained with a close number of samples for both classes. For model training, the pre-trained VGG16 architecture was used with a transfer learning method, and the final layers were retrained and fine-tuned. The model was configured specifically for binary classification. The entire dataset was randomly split into 70% training, 20% validation, and 10% testing. Model performance was evaluated using accuracy, F1-score, sensitivity, and specificity. Results: The model correctly classified 90 out of 94 images in the test dataset, achieving an overall accuracy rate of 96%. For both classes, the precision, recall, and F1-score values were measured in the range of 95% to 96%. Additionally, the Area Under the Curve (AUC) of the ROC curve was calculated as 0.994, and the Average Precision (AP) score was determined to be 0.995. According to the confusion matrix results, only 4 images were misclassified, consisting of 2 false positives and 2 false negatives. Conclusions: The deep learning model demonstrated high accuracy in differentiating zirconia and metal-supported porcelain restorations on panoramic radiographs, suggesting that material-based AI classification may support clinical decision-making in restorative dentistry. Full article

Knowledge Tracing, which estimates how students’ knowledge evolves during interactions with educational content, is a cornerstone of Intelligent Tutoring Systems. While deep learning models achieve superior predictive performance in this task, they lack interpretability, a limitation that is particularly critical in educational contexts. We introduce gTransformer, a new type of grounded Transformer model bridging deep learning performance with intrinsic interpretability through representational grounding. It adds theory-based parameters to input interaction sequences and uses attention mechanisms to transform them into latent representations. These are projected into enriched parameters that incorporate historical learning context while preserving semantics. Validation demonstrates: (1) structural encoding around theoretical concepts (probing selectivity $\Delta R^2>0.5$); (2) semantic alignment; and (3) functional alignment with quantified confidence. Results show that gTransformer achieves predictive performance competitive with state-of-the-art architectures while offering intrinsically interpretable predictions. The trade-off is characterised by a significant Area Under the Curve (AUC) gain over traditional theory-based models (+19.9%), with a minimal cost (3.9%) relative to non-interpretable configurations. Critically, gTransformer enables context-aware personalisation by differentiating students based on longitudinal learning trajectories rather than immediate responses, capturing patterns that traditional models cannot represent. This offers a practical path toward adaptive instruction driven by artificial intelligence that is both accurate and interpretable. Full article

Atopic dermatitis (AD) is a chronic inflammatory skin disorder that is significantly contributed to by epigenetics. We developed a machine learning-based framework to identify DNA methylation biomarkers associated with AD classification and severity. Genome-wide methylation data from peripheral blood were processed using four feature selection algorithms: coarse approximation linear function (CALF), elastic net (EN), minimum redundancy maximum relevance (mRMR), and recursive feature elimination with cross-validation (RFECV). The integrative framework identified a central panel of 8 CpG sites that achieved an area under the curve (AUC) of 1.00 in the test set. This panel demonstrated high disease specificity, showing poor classification performance for systemic lupus erythematosus (AUC = 0.46), Crohn’s disease (AUC = 0.50), and oral squamous cell carcinoma (AUC = 0.58). Severity prediction using RFECV-selected 63 CpG sites (RFE63) achieved high accuracy across classifiers, with Random Forest (accuracy = 0.94) outperforming the others. The functional enrichment of CpG-associated genes highlighted key immune-related transcriptional regulators, including STAT5A, RUNX1, MEIS1, and PAX4. These genes are linked to chromatin remodeling, T helper cell differentiation, and interleukin-2 regulation, which are critical in AD pathogenesis and severity. Our findings demonstrate the utility of machine learning-integrated epigenomics in identifying robust, disease-specific biomarkers for AD diagnosis and monitoring, offering new insights into the molecular mechanisms underlying childhood AD. However, further validation in large-scale independent cohorts is required to confirm their clinical robustness and generalizability. Full article