Search Results (1,460)

Knowledge tracing supports adaptive intelligent tutoring systems by estimating learners’ evolving knowledge states from sequential interactions. However, existing models often rely on dataset-specific correlations and may suffer substantial performance degradation under distribution shifts. This paper proposes CausalSTKT, an SCM-guided knowledge tracing framework that integrates spatiotemporal modeling over a global item–knowledge-component bipartite graph with disentangled representation learning. A relation-aware graph encoder captures high-order dependencies among items and knowledge components, while a temporal transition module models the evolution of learner states. To reduce sensitivity to environmental variation, the learned representation is decomposed into a mastery component and an environment component using a directional decorrelation penalty. An environment-resampling mechanism, which recombines the mastery component of one learner–step pair with the environment component of an unrelated pair drawn at random within the mini-batch, is further introduced to encourage predictions that are stable across environments. We also derive an out-of-distribution risk bound showing that a smaller representation-entanglement residual leads to a tighter generalization bound under the stated assumptions. Experiments on five real-world educational datasets demonstrate that CausalSTKT achieves a mean AUC of 0.8371 and a mean accuracy of 0.8136, exceeding the strongest baseline on each metric by 3.42 and 3.61 percentage points, respectively. In the evaluated zero-shot cross-dataset transfer settings, its average AUC drop is 9.1%, compared with 20.0% and 21.1% for the two baselines evaluated under the same protocol. These results indicate that CausalSTKT provides a robust computational approach to knowledge tracing in intelligent learning systems affected by distribution shifts. Full article

Background: Estimated pulse wave velocity (ePWV) is a non-invasive marker of arterial stiffness with potential relevance for chronic kidney disease (CKD) risk assessment. This study aimed to investigate the association between ePWV and CKD risk in the US and Chinese populations and to evaluate its discriminative performance using machine learning approaches. Methods: Data were obtained from the National Health and Nutrition Examination Survey (NHANES, 2005–2018, weighted n ≈ 100.9 million) and the China Health and Retirement Longitudinal Study (CHARLS, 2011–2015, n = 21,853). In NHANES, CKD was defined according to the 2021 KDIGO criteria as estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m² or a urinary albumin-to-creatinine ratio (ACR) ≥ 30 mg/g; in CHARLS, where urinary albumin was not measured, CKD was defined by the eGFR criterion alone, and this difference in case definition was taken into account when interpreting the results. Logistic regression and restricted cubic spline models were used to examine the association between ePWV and CKD, with subgroup analyses stratified by demographic and clinical characteristics. Multiple machine learning models were developed in NHANES and externally validated in CHARLS; model discrimination was assessed using the area under the receiver operating characteristic curve (AUROC), and feature importance was interpreted using SHapley Additive exPlanations (SHAP) values. Results: Higher ePWV was consistently associated with higher odds of CKD in both cohorts (NHANES: odds ratio [OR] = 1.505 per 1 m/s, 95% confidence interval [CI] = 1.459–1.552; CHARLS: OR = 1.434 per 1 m/s, 95% CI = 1.368–1.503; both p < 0.0001), with dose–response relationships observed. Formal interaction tests confirmed significant effect modification by sex (both cohorts) and by BMI and diabetes (NHANES only). Point estimates were higher in men and in NHANES obese and diabetic subgroups, but interactions across smoking and alcohol strata were not statistically significant. Among the machine learning models evaluated, discrimination was moderate and comparable across algorithms (LightGBM: AUROC = 0.804 in internal validation and 0.793 in external validation), and DeLong tests showed no significant difference between LightGBM and XGBoost (p = 0.0655 and 0.0684, respectively). SHAP analysis identified ePWV as the highest-ranking feature, surpassing uric acid, lipid levels, and diabetes history. Using the Youden index, the optimal ePWV cutoff for identifying CKD was 10.15 m/s in NHANES (sensitivity 0.658, specificity 0.695) and 10.568 m/s in CHARLS (sensitivity 0.658, specificity 0.718). Conclusions: Elevated ePWV is significantly associated with eGFR-defined CKD across the US and Chinese populations studied. These findings support ePWV as a potentially useful marker for CKD risk stratification; however, given the cross-sectional design of both cohorts, its predictive value requires confirmation in prospective studies. It is important to emphasize that no non-invasive calculated metric can replace direct measurement of serum creatinine and urinalysis for identifying individuals at risk of CKD in routine clinical practice. Full article

Background/Objectives: Type 2 diabetes mellitus (T2DM) is characterized by overlapping cardiovascular, renal, metabolic, and hepatic-risk abnormalities. We characterized this integrated phenotype, examined SCORE2-Diabetes gradients, and assessed whether routinely available variables could classify established atherosclerotic cardiovascular disease (ASCVD). Methods: This cross-sectional study included 232 consecutive adults with T2DM. SCORE2-Diabetes tertiles in the full cohort were analyzed descriptively, and a sensitivity analysis was restricted to participants aged 40–69 years without established ASCVD or severe target-organ damage. FIB-4 was recalculated from age, aspartate aminotransferase, alanine aminotransferase, and platelet count. Elastic-net logistic regression, random forest, and gradient boosting were evaluated using nested stratified five-fold cross-validation, with all preprocessing and hyperparameter tuning confined to the training folds. Results: Established ASCVD was present in 49 participants (21.1%), corresponding to 4.45 events per candidate predictor. The SCORE2-Diabetes-eligible sensitivity subgroup included 118 participants (50.9%); 72.9% were in the ≥20% 10-year-risk category. FIB-4 was available for 231 participants (median 1.26 [IQR 0.96–1.80]). Nested cross-validated ROC AUCs were 0.675 (95% CI 0.582–0.763) for elastic-net logistic regression, 0.670 (0.581–0.754) for random forest, and 0.674 (0.589–0.752) for gradient boosting; balanced accuracies were 65.4%, 62.8%, and 56.2%, respectively. Conclusions: The cohort had a high and heterogeneous cardio–renal–metabolic burden. SCORE2-Diabetes findings from the full cohort are descriptive because the score is not intended for patients with established ASCVD or severe target-organ damage. The machine-learning models showed only modest, internally validated discrimination and are not suitable for clinical deployment without larger prospective cohorts and external validation. Full article

Small-radius curved tunneling makes shield machine attitude control especially difficult. Intensified soil-machine interaction under these conditions creates a high risk of snakelike motion, which can compromise both construction safety and segment assembly quality. Accurate advance prediction of attitude parameters is therefore critical for timely course correction. We propose CNN-PatchTST, a hybrid deep learning model that integrates four complementary components. A convolutional neural network (CNN) extracts local temporal features. A PatchTST-based Transformer encoder applies global self-attention over long sequences. A direct mapping branch produces short-range inertial estimates, and a gated fusion layer performs adaptive signal integration. Operating as a unified architecture, the model simultaneously predicts all 12 key attitude parameters five steps ahead, providing operators with roughly five minutes of advance warning. Validated on construction data from the Fangbai Intercity Railway (Guangzhou Metro, minimum curve radius 350 m), CNN-PatchTST achieves a mean coefficient of determination (R²) of 0.992 across all 12 attitude parameters under three independent random seeds. Mean absolute errors (MAE) for the front and rear shield azimuths reach 0.578° and 0.430°, respectively. A pure-inertia baseline (using only historical attitude values) attains R² = 0.988, yet its azimuth MAE is 4.3 times higher than that of the full model. This result confirms that modeling control parameters is essential for accurate angular prediction. SHAP-based sensitivity analysis yields three further insights. First, historical attitude parameters account for approximately 97% of total feature importance. Second, the four most recent steps contribute over 50% of predictive power. Third, a lagged predictive association of roughly 3–4 min is observed between thrust jack pressure differentials and shield tail deviation response, suggesting a potential lagged association that warrants further causal validation. Collectively, these findings demonstrate that CNN-PatchTST delivers accurate and interpretable multi-step attitude predictions, establishing it as a practical tool for on-site guidance during small-radius shield tunneling. Full article

Introduction: Ruptured intracranial aneurysms (RIAs) are a significant cause of morbidity and mortality. While individual-level risk factors for aneurysmal subarachnoid hemorrhage (aSAH) are well established, the influence of meteorological variables has been widely debated but remains unclear. Methods: We retrospectively analyzed 1504 endovascularly treated intracranial aneurysm cases from two stroke centers (2018 to 2024). We matched daily weather data to presentation dates. We used variance inflation factor (VIF) analysis to remove collinear features. Multivariable logistic regression models were adjusted for age and sex. We developed extreme gradient boosting (XGBoost) models using weather variables alone and in combination with demographic covariates (age, sex, race, smoking status, and family history). Results: Of 1504 cases, 377 (25.1%) presented with rupture. In univariate logistic regression, greater humidity (odds ratio [OR] 0.988 per 1% relative humidity, 95% confidence interval [CI] 0.979 to 0.998, p = 0.0135) and greater ultraviolet (UV) index (OR 0.951 per index unit, 95% CI 0.914 to 0.990, p = 0.0146) were associated with reduced odds of rupture, whereas greater snow depth (OR 1.360 per inch, 95% CI 1.068 to 1.731, p = 0.0126) and advanced moon phase (OR 1.589 per full lunar cycle, 95% CI 1.064 to 2.373, p = 0.0237) were associated with increased odds. On multivariable analysis, only female sex remained protective (OR 0.605, 95% CI 0.386 to 0.949, p = 0.0285), and greater sea level pressure trended toward lower odds of rupture without reaching significance (OR 0.962, p = 0.0523). Findings were consistent in a full-cohort, center-adjusted sensitivity analysis. The weather-only XGBoost model yielded an area under the receiver operating characteristic curve (AUC) of 0.560 and a recall of 78%, which improved to an AUC of 0.590 and a recall of 83% after adding demographic variables. SHapley Additive exPlanations (SHAP) analysis identified precipitation and cloud cover as key meteorological features. Conclusions: Several weather variables correlated with rupture risk in univariate analysis, but overall predictive value was limited. Machine learning improved sensitivity while confirming patient features as the dominant contributors. Weather on the day of presentation was not significantly associated with rupture. Full article

Rift Valley fever (RVF) is a vector-borne zoonotic disease whose occurrence is strongly associated with climatic and environmental conditions, making data-driven approaches potentially valuable for epidemiological surveillance and risk assessment. Using a publicly available historical dataset comprising 180,288 monthly observations from geographically defined administrative units across Kenya between 1981 and 2010, this study investigates machine learning (ML) for the retrospective classification of reported RVF occurrence from contemporaneous climatic, environmental, topographic, and seasonal predictors under an extremely imbalanced classification setting. The dataset provides broad geographic coverage across Kenya over a 30-year historical period; however, because the outcome reflects reported events in historical surveillance records, it is not assumed to constitute a formally population-representative national sample or to capture all underlying RVF transmission. Each observation represents a geographic unit and observation month, and the response indicates whether an RVF event was reported during that corresponding period. Therefore, the present analysis should be interpreted as contemporaneous outbreak classification rather than as a fixed-horizon prospective forecast. Thirteen classifiers representing distinct learning paradigms were systematically evaluated: Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Classification and Regression Tree, Naive Bayes, Support Vector Machine, Weighted Logistic Regression, XGBoost, LightGBM, CatBoost, Balanced Random Forest, EasyEnsemble, and RUSBoost. Model performance was assessed before and after SMOTENC-based rebalancing using overall and class-specific metrics, including accuracy, precision, sensitivity, specificity, F1-score, ROC–AUC, and precision–recall-based measures. Under the retrospective stratified hold-out benchmark, XGBoost, CatBoost, Balanced Random Forest, and LightGBM achieved ROC–AUC values of 0.9176, 0.9175, 0.9114, and 0.9062, respectively. Balanced Random Forest attained the highest outbreak sensitivity (0.8851), although at the cost of very low precision, illustrating that high rare-event detection can generate a substantial false-alert burden in surveillance settings. SMOTENC produced strongly model-dependent effects: it increased outbreak sensitivity for XGBoost, LightGBM, CatBoost, KNN, CART, and RUSBoost, but substantially reduced sensitivity for Balanced Random Forest and EasyEnsemble. SHAP-based interpretability analysis indicated that month, rainfall, and slope were among the most influential predictors and further showed that class rebalancing can alter the distribution of feature contributions. Overall, the findings demonstrate that modeling reported RVF occurrence under severe class imbalance requires joint evaluation of minority-class detection, false-positive behavior, discrimination, and model interpretability rather than overall accuracy alone. The present results establish a retrospective classification benchmark for climate-informed RVF risk assessment, but they should not be interpreted as an autonomous outbreak-warning system. Translation into prospective early-warning prediction will require an explicit forecasting horizon, predictors constructed exclusively from information available before the target period, temporally and geographically independent validation, and decision thresholds evaluated against an operationally acceptable false-alert burden. Full article

In the Interval Job Shop Problem (IJSP), operation processing times are represented by intervals, making machine-sequencing decisions sensitive to temporal uncertainty. We propose Risk-Aware Partial-Order Reinforcement Learning with Timed Automata Completion (RAPO-RL-TAC), which couples Risk-Aware Partial-Order Reinforcement Learning (RAPO-RL) for machine-order construction with Timed Automata Completion (TAC) for execution-time resolution of deferred sequencing decisions. Risk awareness focuses on preserving temporal flexibility in machine-order relations whose preferred ordering is sensitive to interval uncertainty. RAPO-RL selectively commits comparatively determinate machine conflicts while retaining a bounded set of timing-sensitive relations. TAC completes unresolved relations as execution evolves, while statistical model checking characterises completion-time variability across timed executions. On 14 ORB and LA benchmark instances, RAPO-RL-TAC achieves mean midpoint makespans 2.92% and 1.63% lower than population-based neighbourhood search (PNS) and genetic algorithm (GA), respectively. Compared with reproduced Fast Elitist Artificial Bee Colony (fEABC) variants, RAPO-RL-TAC achieves a lower midpoint than at least one variant on 7 of 14 instances. In the controlled ablation study, RAPO-RL-TAC achieves a 13.85% lower mean midpoint makespan than hard enforcement of the learned relations. These results indicate that risk-aware selective commitment preserves temporal flexibility while maintaining competitive nominal schedule quality under interval uncertainty. Full article

Background: Deep-learning artificial intelligence (AI) for mammographic screening moved from proof of concept to randomised evaluation in a single decade. We reviewed and meta-analysed its diagnostic accuracy and its effect on screening programmes, covering the literature published between 2015 and 2025. Methods: PubMed and Europe PMC were searched from 1 January 2015 to 31 December 2025, supplemented by ClinicalTrials.gov and by forward and backward citation searching (PRISMA 2020, PRISMA-DTA, PRISMA-S). Eligible studies evaluated a deep-learning system for cancer detection or triage in a screening population against a histopathological reference standard. Four syntheses were performed: standalone accuracy pooled on the logit-AUC scale (A), a bivariate sensitivity–specificity model (A2), cancer detection rate ratio for AI-integrated versus standard reading (B), and recall rate ratio (C). Random-effects models used restricted maximum likelihood with Knapp–Hartung intervals. Risk of bias was assessed with QUADAS-2 and QUADAS-C, and certainty with GRADE. Results: Twenty-six studies (27 reports, 2019–2025) were included. Pooled standalone AUC across 14 studies and 1,214,885 examinations was 0.890 (95% CI 0.858–0.915), with I² = 97.4% and a 95% prediction interval of 0.731–0.960. Neither publication year (p = 0.79) nor enriched versus consecutive sampling (p = 0.94) explained this dispersion in meta-regression. The bivariate model (k = 9) gave a summary sensitivity of 73.3% (64.5–80.6) at a specificity of 92.4% (86.8–95.8). Across randomised and paired prospective trials (k = 3), the pooled detection rate ratio was 1.13 (0.83–1.55), with the MASAI randomised trial alone reporting 1.29 (1.09–1.51) and a 44% reduction in screen reading. Non-randomised implementation studies (k = 5), which include a 463,094-women German programme evaluation, pooled to 1.22 (1.08–1.37) with little dispersion (I² = 19.0%). Recall changed little overall (0.95, 0.81–1.12). Certainty was very low for accuracy outcomes and moderate for the single randomised trial. Conclusions: A decade of evidence supports AI as a second reader and triage tool in organised screening, not as an autonomous replacement for the radiologist. Pooled accuracy is high on average but so dispersed that it cannot be transferred to a new programme; local validation before deployment remains necessary. The larger and more precise detection gains come from non-randomised designs, which is the pattern confounding would produce, so the randomised evidence remains the anchor. Interval-cancer and mortality endpoints are still awaited. Full article

Background and Objective: Obstructive sleep apnea (OSA) is a fatal widespread sleep-related breathing disorder and a major risk factor for cardiovascular and cerebrovascular diseases, significantly impacting older adults’ health worldwide. Due to the risk of such complications, timely and accurate identification of OSA is crucial. Polysomnography is considered the most accurate technique for detecting OSA; however, it is limited by its complexity and multi-channel requirements. A promising alternative is electrocardiogram (ECG)-based diagnosis, which continuously monitors heart rhythm and captures subtle cardiac changes associated with OSA. Nevertheless, existing ECG-based approaches still face challenges related to complex feature engineering, limited capture of complementary temporal–spectral information and global dependencies, along with inadequate feature recalibration and fusion, which can restrict OSA detection. Thus, further improvements are still required to achieve clinically reliable performance. Methods: To address these challenges, this study proposes SApneaNet, a novel advanced deep learning method for detecting OSA events using ECG signals. The proposed approach employs the continuous wavelet transform (CWT) to convert ECG signals into RGB log-scalograms, enabling the simultaneous analysis of temporal and frequency-domain features. The generated RGB log-scalograms are then fed into a deep CNN encoder with adaptive squeeze-and-excitation (ASE), followed by a transformer and an adaptive gated feature fusion (AGFF) architecture. In this framework, to improve OSA detection performance, the CNN extracts rich local features, the ASE module performs channel-wise recalibration to enhance feature representations, the transformer performs data-parallel processing and captures global contextual dependencies, and the AGFF mechanism adaptively emphasizes informative features while suppressing less relevant ones. Results: The experimental results on the Apnea-ECG dataset showed that the model achieved a sensitivity of 94.7%, specificity of 95.2%, F1-score of 93.5%, accuracy of 95.1%, Cohen’s kappa of 89.4%, and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.989 for per-segment classification. Furthermore, for per-recording classification, the model achieved an accuracy of 100.0%, a mean absolute error (MAE) of 2.025, and a Pearson correlation coefficient (PCC) of 0.992. Overall, the experimental results demonstrated that the proposed model achieved excellent and competitive performance compared with other advanced state-of-the-art methods for OSA classification. Conclusions: The proposed model demonstrates strong efficacy in OSA detection, providing a novel and robust alternative to conventional diagnostic methods. The model’s reliable and consistent diagnostic performance highlights its potential for integration into practical OSA diagnostic systems, including home-based health monitoring devices and clinical decision-support tools. Full article

With the rapid proliferation of collaborative mobile applications, edge computing has emerged as a promising paradigm to deliver ultra-low latency and localized services. However, conventional Edge Server Placement (ESP) strategies primarily optimize physical transmission delay and server load balancing, largely overlooking the intense cross-server communication overhead induced by user social interactions. Furthermore, directly utilizing fine-grained user trajectories and social graphs to guide placement poses severe privacy risks. In this paper, we propose a socially aware edge server placement framework, termed Fed-STSR, where the primary contribution is a multi-objective optimization formulation that explicitly integrates user social relationships alongside network latency, service migration, and server load balancing. To enable this framework without compromising user confidentiality, federated representation learning coupled with a calibrated Laplace differential privacy mechanism serves as an enabling layer, compressing sensitive spatiotemporal patterns into bounded representations locally. The resulting discrete placement problem is solved using an enhanced Trees Social Relations (TSR) algorithm operating over sparse adjacency structures. Extensive trace-driven simulations based on the Gowalla and Brightkite datasets mapped onto real urban base station layouts demonstrate the effectiveness of the proposed approach. Compared to state-of-the-art baselines, Fed-STSR reduces average service latency by up to 31.5% and achieves superior server load balance while maintaining rigorous user-level privacy guarantees. Full article

Background: This study aimed to evaluate the performance of machine learning models in predicting depression risk among older adults living alone and to identify the features contributing to those predictions using explainable artificial intelligence (XAI). Methods: We analysed 2022 nationwide survey data in Korea. A total of 1007 older adults remained after excluding respondents who lived in multi-person households, were aged < 65 years, or had physician-diagnosed dementia. Depression risk was defined using the CES-D-10 (cutoff ≥ 10). After removing features with high multicollinearity, logistic LASSO selected 23 predictors. Six algorithms were fitted using the training set, with hyperparameter tuning performed by 5-fold cross-validation where applicable, and evaluated in a held-out test set following a 70/30 split. SMOTE was applied only to the training data. Performance was summarised using AUC, sensitivity, specificity and the F1 score with bootstrap 95% confidence intervals, and stability was assessed by repeated stratified cross-validation. SHAP values provided explainability. Results: LightGBM achieved an AUC of 0.802 (95% CI 0.747–0.852), followed by Random Forest (0.794) and Logistic Regression (0.779). These differences were small relative to the uncertainty of the estimates. SHAP analysis identified oral health-related quality of life, satisfaction with relationships with children, frequency of social contact, overall life satisfaction, satisfaction with health status, and age as the most influential features. IADL limitations, diabetes, hypertension, and perceived social class contributed to predictions with smaller effects. Conclusions: An explainable LightGBM model achieved an AUC of 0.802 for depression risk among older adults living alone and identified psychosocial and health-related features, particularly oral health and social connectedness, that may help inform future screening strategies. Full article

Stock market volatility forecasting is important for risk management, portfolio allocation, and investment decision-making. This study provides a bibliometric and methodological review of stock market volatility measurement and forecasting research published between 2016 and 2025. A model-neutral search of the Web of Science Core Collection, conducted on 18 August 2026, identified 440 records. Following title, abstract, full-text, and document-type screening, 177 eligible journal articles were retained. To assess search-term sensitivity, a supplementary search conducted on 4 September 2026 using alternative volatility terminology identified 33 additional eligible studies, yielding a final corpus of 210 studies. Bibliometrix/Biblioshiny and structured methodological classification were used to examine the field. Econometric approaches remained dominant (168 studies; 80.0%), followed by Machine Learning (25; 11.9%), Deep Learning (8; 3.8%), and Hybrid approaches (9; 4.3%). The evidence reveals substantial methodological diversification beyond conventional GARCH models and increasing use of realized and implied volatility, high-frequency information, sentiment, macroeconomic variables, and uncertainty indicators. No methodological family demonstrates universal forecasting superiority, as performance depends on markets, horizons, information sets, benchmarks, and evaluation criteria. Overall, the literature reflects methodological diversification, information enrichment, and selective integration rather than replacement of econometric models by artificial intelligence. Although the review is limited to the Web of Science Core Collection, the sensitivity analysis demonstrates the importance of alternative terminology in identifying relevant studies. Full article

Over the past decade, the Punjab plains of Northern India have experienced recurrent flooding driven by hydroclimatic variability, specifically shifts in western disturbances that have intensified monsoon precipitation. Following a devastating flood in 2025, the region remains highly vulnerable to hydrological extremes, a risk further exacerbated by shifting land use and agricultural patterns, geomorphic parameters, and complex fluvial systems. This study assesses flood susceptibility by integrating multi-sensor satellite observations, multi-temporal Sentinel-1 backscatter signals, and refined runoff potential estimates derived from local climate zones, accounting for land cover, soil type, and infiltration characteristics, into machine learning frameworks. The model is trained using a 2025 flood inventory generated from a synthetic aperture radar backscatter threshold ratio. The calibrated frameworks are applied to the 2023 flood events to test independent transferability. The temporal consistency and predictive performance of the models are evaluated using the precision–recall trade-offs, threshold-dependent predicted probability distribution, and Shapley Additive exPlanations (SHAP). Results indicate more balanced classification performance of Random Forest and Extreme Gradient Boosting in comparison to Artificial Neural Network performance that exhibits higher recall with lower precision. The model performance for 2023 models is considered more robust, with greater class separability of 2023 flood events than for 2025. Probability distributions for both events further demonstrate model-dependent threshold behavior, highlighting a trade-off between flood detection sensitivity. SHAP identifies rainfall, soil moisture, runoff, and elevation as the dominant contributors. The analysis further indicates that all model frameworks effectively capture the physical control of hydrological and topographical variability on the temporal flood events. The consistent contribution of hydrological and topographical factors across the two events supports model transferability, while threshold sensitivity, uncertainty, and spatial dependence are important considerations for flood susceptibility modelling. The results reflect a balanced interaction between extreme rainfall, runoff potential, and topographic control in causing periodic floods in the Punjab plains. Full article

Financial markets exhibit complex nonlinear dynamics driven by interactions between firm-specific characteristics and macroeconomic conditions, requiring robust mathematical models capable of capturing structural changes and temporal heterogeneity. This study proposes an explainable machine learning framework that combines regime-switching analysis and predictive modeling to investigate the long-term resilience of firms listed in the BIST 100 Index. Structural breaks in monthly return and volatility series are first detected using the Pruned Exact Linear Time (PELT) algorithm, enabling the classification of firm trajectories into resilient and fragile market regimes. A Regime Persistence Score is then introduced to quantify the proportion of time each firm remains in the resilient state. This score serves as the response variable in a Random Forest model that evaluates the influence of firm-specific financial indicators and macroeconomic variables, including exchange rates, producer price inflation, and commercial loan interest rates. In addition, a forward-looking classification model estimates the probability that firms will transition into a fragile regime within the subsequent six months. The proposed framework achieves an AUC of 0.706 and demonstrates stable predictive performance under alternative regime definitions, penalty parameters, cost functions, and cross-validation strategies. Explainability analysis derived from Shapley Additive Explanations (SHAP) data shows book-to-market ratio and sensitivities to inflation and interest rate changes as the most important components of enduring resilience. The proposed methodology provides an interpretable mathematical framework for regime detection, nonlinear time-series modeling, and decision support in sustainable financial systems, offering practical value for risk assessment and resilience-oriented portfolio management. Full article

Handwriting analysis has recently emerged as a modality for the early identification of learning disabilities, as it can reveal characteristic patterns associated with motor control, spatial planning, and cognitive processing. Nevertheless, most existing automated handwriting analysis systems focus primarily on Latin scripts, while Arabic handwriting remains largely unexplored despite its distinct structural and morphological characteristics. This research presents a Convolutional Neural Network (CNN) model for the early risk screening of dysgraphia in the handwriting of Arabic-speaking children. A specialist-annotated dataset comprising 259 handwriting samples from fourth-grade students in Saudi Arabia was collected and preprocessed using a pipeline that included grayscale conversion, Otsu binarization, spatial normalization, and data augmentation. A custom CNN architecture with batch normalization and global average pooling (GAP) was trained with focal loss to address class imbalance. Furthermore, ten handwriting features grounded in prior literature were extracted and evaluated using four classical machine learning classifiers under identical experimental conditions. The proposed CNN model achieved satisfactory performance, reaching 88.8% accuracy, 92.3% balanced accuracy, and 98.1% sensitivity, substantially outperforming feature-based machine learning approaches, whose best balanced accuracy reached 62.6%. These results demonstrate that discriminative patterns in Arabic-speaking children’s handwriting can potentially be learned by deep neural networks. The proposed approach could support the early identification of children showing indicators of potential dysgraphia, assisting educators and specialists in making timely referrals for further assessment. Full article