Search Results (67,845)

Tax administrations increasingly use data-driven risk models to prioritize collection resources, yet machine learning applications to personal income tax (PIT) arrears remain limited. This study introduces TaxMind (v1.0), an interpretable risk management framework for predicting whether PIT obligations will progress to mandatory collection and for integrating debtor- and debt-related information into risk-based segmentation. Administrative records from the Lebanese Tax Administration were analyzed; after removing 5385 exact duplicates from 16,010 records, the final dataset contained 10,625 obligations across 9264 taxpayers. Five tuned classifiers were evaluated using a taxpayer-grouped train/test design, with SHAP used for model interpretation. XGBoost achieved the highest observed discrimination (ROC-AUC = 0.784; accuracy = 0.708; F1-score = 0.704), closely followed by Random Forest (ROC-AUC = 0.780); the leading tree-based models substantially outperformed logistic regression benchmarks. SHAP identified Total Tax Amount, Tax Category 2, and Age as the three leading individual encoded features, while the Age contribution was nonlinear and varied across Tax Categories. The principal SHAP ranking was highly consistent under Random Forest. Top-decile ranking by predicted probability captured only 0.4% of the monetary exposure of realized mandatory collection cases, versus 92.7% under probability-weighted exposure ranking. These findings show that debt-related information remains central, but selected debtor characteristics add predictive value, supporting TaxMind as a model-agnostic early-warning framework for preventive tax debt management. Full article

Preeclampsia remains one of the leading causes of maternal and perinatal morbidity and mortality worldwide. Although current first-trimester screening strategies have improved risk assessment, their predictive performance remains limited by the biological complexity and heterogeneity of the disease. Artificial intelligence (AI) has emerged as a promising approach capable of integrating multidimensional clinical and biological data to improve early prediction. This review aims to summarize current evidence regarding AI-based prediction models for preeclampsia, compare their performance with conventional screening strategies, and discuss future directions for clinical implementation. A narrative review of published studies evaluating machine learning and deep learning models for first-trimester prediction of preeclampsia was performed. Studies incorporating maternal characteristics, hemodynamic variables, biochemical biomarkers, imaging, radiomics, and multi-omics data were reviewed. Diagnostic performance, predictor variables, and validation strategies were critically compared. Several studies have reported improved predictive performance of AI models compared with conventional statistical approaches, particularly when multimodal datasets were incorporated. High-performing models achieved area under the receiver operating characteristic curve (AUC) values ranging from 0.84 to 0.92. Across studies, maternal clinical characteristics, mean arterial pressure, uterine artery pulsatility index, placental growth factor, and pregnancy-associated plasma protein-A were the most consistently identified predictors. However, direct comparisons remain limited by methodological heterogeneity. Emerging approaches incorporating inflammatory biomarkers, cell-free nucleic acids, radiomics, and multi-omics technologies showed encouraging results but currently lack sufficient prospective multicenter validation for routine clinical implementation. AI has considerable potential to improve first-trimester prediction of preeclampsia, although prospective multicenter validation, standardized reporting, and implementation studies remain necessary before routine clinical adoption. Future research should prioritize prospective multicenter validation, standardized data collection, explainable AI, and seamless integration into clinical workflows to facilitate implementation in precision obstetric care. Full article

Continuous structural health monitoring is essential for ensuring the safe operation of critical engineering systems. Ultrasonic guided waves are widely used for damage detection and localization due to their ability to cover large areas with high sensitivity to structural changes. This work evaluates the robustness of a previously proposed guided-wave localization approach by applying it to the complex geometry of carbon fiber composite plates with integrated omega stringers. The measurement data used in this study were provided within the framework of the Open Guided Waves project. We employ an interpretable machine-learning framework based on the normalized root mean square deviation to extract damage-sensitive features. Damage localization is further improved by modeling the spatial damage probability using a normal distribution, which enhances spatial coverage of the structure. The influence of 13 damages with progressively increasing size on classification and localization performance is systematically analyzed. The proposed method achieves an average classification accuracy of 95% and a mean localization error of 5 mm, demonstrating its suitability for damage characterization in complex composite structures. Full article

Remaining useful life (RUL) prediction of power transformers is essential for ensuring reliable operation and supporting condition-based maintenance. However, the complex nonlinear degradation characteristics and long-term temporal dependencies of transformer monitoring data pose significant challenges for accurate RUL prediction. Moreover, most existing deep learning-based prediction methods provide deterministic predictions without explicitly quantifying predictive uncertainty, while the consequences of RUL overestimation and underestimation in practical maintenance are inherently asymmetric. Therefore, this paper proposes an uncertainty-aware RUL prediction method for power transformers by integrating the Patch Time Series Transformer (PatchTST) with deep evidential regression (ER-PatchTST). PatchTST is used to transform long time-series inputs into patch-level representations, enabling the model to capture local temporal patterns and long-range dependencies through patch-wise tokenization and channel-independent modeling. In addition, the deep evidential regression module is designed by placing a Normal–Inverse-Gamma (NIG) prior over the parameters of the Gaussian likelihood, which can simultaneously predict RUL and quantify aleatoric and epistemic uncertainties in a single forward pass. Furthermore, a safety-oriented RUL indicator and hierarchical warning strategy are developed, and different maintenance actions are initiated when alarms at different levels are triggered. Experiments on the ETT dataset demonstrate that ER-PatchTST achieves competitive forecasting performance compared with state-of-the-art time-series forecasting methods while simultaneously providing predictive uncertainty estimates. On the DGA dataset, ER-PatchTST achieves the best RUL prediction performance among the compared methods and provides informative uncertainty quantification for condition-based maintenance decision support. Full article

This research analyzes the association between the quality of virtual learning environments and students’ digital literacy in higher education. From a hierarchical perspective, virtual learning environment quality was defined as a second-order emergent construct integrating technological usability and accessibility with pedagogical mediation. Similarly, student digital literacy was conceived as a second-order emergent construct encompassing information and digital resource management, digital content production, digital communication, collaboration and citizenship, as well as problem-solving and digital autonomy. The study followed a quantitative, non-experimental and cross-sectional design. Data were obtained from 342 university students enrolled in the 2026-I academic term and examined through partial least squares structural equation modeling, applying a two-stage procedure in ADANCO. First-order constructs were estimated as reflective Mode A consistent variables, whose standardized latent scores served as indicators for second-order Mode B emergent constructs. Findings confirmed significant external weights and adequate collinearity levels across all second-order components. The structural model showed a positive and statistically significant association between virtual learning environment quality and student digital literacy (β = 0.795, p < 0.001), with VLEQ accounting for 63.2% of the variance in SDL within the estimated model. Overall, the results suggest that higher perceived digital literacy is associated not only with platform access, but also with pedagogically mediated environments that support interaction, autonomy, and responsible digital engagement. Full article

Digital artifacts are preferred by both students and instructors to achieve various educational goals; yet the understanding of how digital artifacts and their properties are associated with academic development remains insufficient. Such understanding is also important as the design, modification, and use of digital artifacts, such as infographics, videos and illustrations, support both learners and instructors. Thus, drawing on the theory of semiotic mediation, this study examines how three core properties of digital artifacts, i.e., design, elaborateness, and semantic ability, are associated with learner outcomes through self-regulated learning, student participation, individual innovativeness, and learning continuance. Digital artifacts are conceptualized broadly to include conventional digital resources. Data were collected from 350 university students in Saudi Arabia and analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM). The findings show that design, elaborateness, and semantic ability are positively associated with both self-regulated learning and student participation. Such understanding, in turn, is associated with individual innovativeness, which is positively associated with learning continuance and learner performance. The results suggest that the educational value of digital artifacts depends not only on their technological availability but also on how their design characteristics support learners’ regulation, participation, experimentation, and sustained engagement. The study extends the application of semiotic mediation theory to contemporary digital learning environments and provides a behavioral explanation of how digital artifact properties are associated with academic development. Full article

Temperate native grasslands in southeastern Australia have been extensively cleared for agriculture, and the remaining patches are under growing pressure from further land use change, climate variability, and invasive species. Mapping and monitoring their distribution and the cover of native and exotic grasses are critical for their conservation and management. Field-based methods are not always scalable or time-effective, and this study aimed to develop a scalable method to map and monitor the fractional cover-class maps of native C3 and native C4 grass cover as a component of remnant native grasslands on the western outskirts of Melbourne, Victoria, Australia. Field-based reference data for training and validation of random forest machine learning models were collected across multiple sites in 2021. Sentinel-2 optical spectral bands and vegetation indices were used as the primary input data, and Sentinel-1 Synthetic Aperture Radar (SAR)-derived Grey Level Co-occurrence Matrix (GLCM) texture metrics were assessed for their capacity to improve the model. Results show that random forest models trained on Sentinel-2 data without GLCM texture information derived from Sentinel-1 SAR data provided a moderate overall accuracy (C3: 59.1%, C4: 78.1%). Class-specific metrics showed that reliability was highest for better represented lower-cover classes, particularly the 6–25% native C3 class and the 0–5% native C4 class, while higher-cover classes were less reliable because of the limited number of training and validation samples. Grass cover fractions were modelled well for sparse to moderate grass cover, but dense grass cover was not modelled accurately, probably due to limited high-cover samples in the training dataset. Model performance was not improved by the inclusion of Sentinel-1 SAR-derived GLCM texture metrics, indicating that C-band VH-polarised SAR is not sensitive to the fine-scale structural heterogeneity that characterises native grassland ecosystems. Sparse native C3 and C4 grasses could be mapped most reliably in the lower-cover classes as a component of grasslands with optical remote sensing, and the method developed here can now be applied to enable evidence-based management of grasslands, biodiversity conservation and the monitoring of grassland composition in the WGR and elsewhere. Higher-resolution structural datasets and more sophisticated machine learning approaches may be required to accurately predict native C3 and C4 grass cover fractions in denser grasslands. Full article

Accurate remaining useful life (RUL) prediction of air circuit breakers (ACBs) is crucial for condition-based maintenance. However, existing data-driven prognostic methods suffer from electromechanical feature fragmentation, cross-device domain shifts, and the inability to penalize safety-critical late predictions. This study proposes an explainable RUL prediction framework via physics-informed feature fusion. Through full-lifecycle monitoring, novel indicators, including the electromechanical coupled degradation index (EMCDI) and the contact spring over-travel consumption rate (CSOCR), are introduced to decode interactive degradation cycles. To eliminate the interferences of initial manufacturing tolerances, a phase-decoupled normalization strategy empowers a random forest (RF) model to achieve cross-device transferability in a two-device proof-of-concept experiment, requiring only 50 initial operations for target calibration. Additionally, a safety-oriented asymmetric penalty score (APS) is integrated into the evaluation framework to explicitly penalize hazardous life overestimations. Experimental results demonstrate a full-lifecycle R² of 0.9936 and a mean absolute error (MAE) of 38.5136. While the early-stage R² of 0.6880 objectively reflects the statistical flatness of the equipment’s healthy plateau, the framework maintains robust tracking capabilities across the entire lifespan, surpassing mainstream deep learning algorithms such as CNN, MLP, and LSTM. The proposed method consistently achieves a conservative, risk-averse predictive distribution for industrial reliability. Finally, model-level permutation importance analysis confirms that the RF model prioritizes physics-informed indicators rather than relying on spurious curve fitting. Full article

Reliable streamflow forecasting with quantified uncertainty is essential for water resource management, flood mitigation, and climate adaptation in semi-arid regions. This research introduces a framework that combines three conformal prediction techniques, including Split Conformal Prediction (SplitCP), Cross Validation Plus (CV+), and conformal quantile regression, with two probabilistic machine learning algorithms, namely Natural Gradient Boosting (NGBoost) and Probabilistic Gradient Boosting Machines (PGBM), to quantify uncertainty in hydrological modeling of the Sattarkhan Dam in East Azerbaijan Province, located in north-eastern Iran. The data collected ranged from 21 March 1996 to 22 September 2022 and were divided into two chronological groups: training (70%) and testing (30%) for modeling. Probabilistic prediction quality was evaluated using the continuous ranked probability score (CRPS) and negative log-likelihood (NLL). In contrast, for conformal prediction, we used the mean predicted interval width, effective coverage, and coverage width criteria. Results in terms of correlation coefficient (CC), mean absolute error (MAE), and root mean square error (RMSE) with the test dataset suggest improved performance by NGBoost (RMSE: 0.833 m³/s, CC: 0.918, MAE: 0.375) using optimal values of user-defined parameters in comparison to PGBM (RMSE: 0.909 m³/s, CC: 0.902, MAE: 0.388). NGBoost outperforms PGBM in probabilistic prediction. Its higher coverage indicates CV+ as the most effective uncertainty estimation method for this dataset. These findings support model reliability and inform future decision-making. These findings support operational forecasting and risk-informed decision-making in semi-arid regions. Also, the framework provides a transferable template for similar hydrological uncertainty studies. Full article

Background: Knee MRI is one of the most performed MR exams in outpatient settings and has a crucial role in detecting pathology. To increase access to advanced imaging and provide valuable diagnostic information to the clinician, reducing imaging acquisition time while preserving good image qualities is crucial. Deep learning-based reconstruction (DLR) algorithm (SubtleMR™) may be an alternative to conventional MRI with reduced scanning time while preserving image quality and diagnostic value. Purpose:To compare image quality and diagnostic performance among conventional sequences and an automated accelerated deep learning-based reconstruction (DLR) knee algorithm method in patients with knee pain. Material and Methods: A total of 50 consecutive patients (26 men, mean age of 43.5 ± 10 years) with knee pain who underwent conventional (acquisition time: 13 min and 27 s) and accelerated DLR (acquisition time: 6 min and 45 s) knee MRI between November 2020 and December 2020 on a 1.5 T MRI system were prospectively enrolled. Two independent musculoskeletal radiologists compared the image quality artifacts and pathology using a 3-point Likert scale (1–3). Data were compared using paired Wilcoxon signed rank tests.Results:Forty-three patients with 86 knee MRI examinations were included. A comparison of conventional sequences and accelerated DLR by both readers revealed that there were no significant differences in overall diagnosis, image quality, and artifacts. Accelerated DLR knee MRI is comparable with conventional acquisitions, with an average time difference of 6 min and 42 s per patient (49.81%). Conclusions: Accelerated DLR knee MRI substantially reduces acquisition time (~49.81%) while maintaining equivalent diagnosis and image quality. Full article

Catastrophic forgetting is a major challenge for deep learning models when they are incrementally trained on a sequence of new data. Reducing this forgetting in image and video data has been the primary research focus, but less attention has been given to textual domains, where discrete token distributions and semantic shifts occur across different topics. Furthermore, standalone strategies proposed for reducing catastrophic forgetting still have room for improvement. To this end, this paper proposes a synergy of stratified memory replay with parameter regularization for a BiLSTM-based incremental learning model to mitigate catastrophic forgetting in sequential text datasets. The stratified replay mechanism replays a small buffer of historical data samples into current training phases to preserve the old data patterns, while the parameter regularization penalizes modifications to the neural network weights crucial to the past tasks. The proposed approach is evaluated in an incremental training pipeline on distinct textual datasets, including software bug reports (Task A), a news dataset (Task B), and emails (Task C). The evaluation results demonstrate that the baseline neural network experiences catastrophic forgetting as its initial dataset (Task A) accuracy drops from $87.53\%$ to $11.75\%$. The standalone experience replay approach manages to retain Task A accuracy at $80.01\%$, down from its peak of $86.56\%$, while for Task B, it achieves $93.97\%$, down from the peak of $98.37\%$. The buffer sensitivity analysis indicates the model accuracy improves with increasing replay buffer size. The parameter regularization approach effectively reduces catastrophic forgetting, but it remains less effective for disruptive text distribution sequences, resulting in noticeable forgetting on prior tasks and reduced plasticity on later tasks. Evaluations of this approach show that Task A accuracy is reduced from $87.87\%$ to $71.09\%$ after training on Task C. The proposed hybrid approach reduces forgetting and preserves Task A and Task B accuracies at $83.73\%$ and $95.04\%$, respectively. Thus, the empirical evaluations demonstrate that the proposed approach is effective and outperforms the standalone experience replay strategy and parameter regularization, limiting the forgetting on the earliest task to just $4.65\%$ compared to $6.55\%$ forgetting of the replay-based method, while allowing enough plasticity for the final task to reach $98.92\%$ accuracy. Full article

Lung cancer remains the leading cause of cancer-related death globally, despite significant advances in diagnosis and treatment. Single biomarker approaches used clinically, such as programmed death ligand-1 (PD-L1) expression levels, have limited capacity for predicting treatment response. Multimodal data analysis using artificial intelligence (AI) offers an innovative scope to integrate diverse data sources—including radiologic imaging, digital pathology, genomics, immunohistochemistry, and Cell Painting morphology—to improve clinical predictions. This review aims to examine multimodal AI applications across the lung cancer treatment landscape related to such data sources. We analyze technical architectures spanning convolutional neural networks for imaging, vision transformers for pathology, and graph neural networks for genomics. We discuss how integrating and learning from heterogeneous data sources requires cross-attention fusion mechanisms. We further analyze critical studies demonstrating that multimodal AI clinical applications achieve superior predictive performance compared to unimodal biomarker methods. Multimodal AI models can augment clinicians in treatment selection, longitudinal monitoring using circulating tumor DNA (ctDNA), and variant interpretation through morphological profiling. We propose developing a multimodal AI model to optimize precision oncology for lung cancer. Full article

The sponge construction serves as a fundamental design framework for hash functions and authenticated encryption algorithms. In differential cryptanalysis of large-state permutations underlying such algorithms, conventional approaches are constrained by state size and diffusion speed. The feature extraction capability of deep learning offers a potential alternative to mitigate these limitations. To improve the distinguishing performance of differential neural distinguishers against sponge-based algorithms, a methodology integrating data construction and network architecture optimization is proposed. Specifically, a multi-sample triplet input format is designed to preserve differential characteristics, and a convolutional block attention module is introduced to capture long-range dependencies along both the channel and spatial dimensions within the large-state permutation. Experimental evaluations were conducted on the Keccak and Ascon algorithms. For Keccak, the maximum distinguishable round number was identified as 3. At this round number, Keccak-p achieved full distinguishability (100% accuracy), while the sponge-based SHA-3-256 attained a distinguishing accuracy of 99.99%, improving upon the previous best result by 0.95 percentage points. For Ascon, the maximum distinguishable round number was 4, where Ascon-p achieved an accuracy of 54.85% with 64 sample pairs—the highest reported accuracy for this setting—while delivering comparable performance at the matched 32-pair setting (53.40% vs. 53.54% in prior work) with approximately one-twelfth of the training epochs; under the same setting, the sponge-based Ascon-Hash256 achieved an accuracy of 53.06%. These findings demonstrate the effectiveness of the proposed framework in enhancing neural distinguisher accuracy against sponge-based algorithms and offer an analytical approach for empirical security evaluation, with results qualitatively consistent with the indifferentiability bound of the sponge construction. Full article

Understanding the spatial distribution patterns of pelagic species such as yellowfin tuna (Thunnus albacares) is essential for ecosystem-based fisheries management. However, characterizing CPUE–environment relationships remain challenging because these relationships may be nonlinear and spatially heterogeneous across large oceanic regions. To address these challenges, we developed an interpretable spatial modeling framework, geographically neural network weighted regression integrated with GeoShapley analysis (GNNWR-GeoShapley), which combines the nonlinear learning capability of neural networks with spatially explicit characterization and interpretation of model relationships. Using Pacific longline fishery data and multi-source environmental variables from 2004 to 2023, we constructed quarterly models of CPUE–environment relationships and compared the performance of GNNWR with Generalized additive model (GAM), geographically weighted regression (GWR), graph neural network (GNN) models, and Geographical Random Forest (GRF). The results demonstrated that GNNWR showed the best overall performance across seasons, effectively capturing nonlinear relationships and spatial heterogeneity in yellowfin tuna nominal CPUE. GeoShapley analysis further revealed that sea surface and subsurface (150 m) temperature and salinity were among the most important environmental variables associated with nominal CPUE variations. Nonlinear response patterns indicated that SST values above approximately 25 °C and T150 values above approximately 19 °C were associated with positive model contributions, whereas higher salinity values (>35) exhibited negative contributions. Moreover, spatial effects represented by the geographical location variable (GEO) and their interactions with environmental variables revealed pronounced spatial heterogeneity, with the contribution patterns of environmental factors varying across seasons and regions. This study provides an interpretable spatial modeling framework for characterizing complex species–environment relationships and offers new insights into the spatial variability of Pacific yellowfin tuna nominal CPUE for fisheries oceanography and sustainable resource management. Full article

Mining and metallurgical wastes contain potentially valuable mineral resources but pose growing environmental risks. This review proposes a mineralogy-guided framework for converting these wastes into low-carbon binders and construction materials. Rather than classifying residues by origin or bulk oxide composition, it distinguishes them according to the reaction roles of their mineral phases. The framework first distinguishes self-sufficient systems, in which activation unlocks the intrinsic mineral inventory of the waste sufficiently for matrix or product formation, from mineralogically compensated systems, in which additional functional mineral solids are required to supply deficient reactive Si, Al, Ca, sulfate, alkalinity, or other phase-forming components. Binary and multicomponent formulations are then interpreted according to the specific mineralogical functions supplied by the complementary solids and the resulting changes in reaction pathways and products. Thermal, hydrothermal, mechanochemical, alkaline, and carbonation-based processing routes are compared in relation to phase composition, co-precursor function, and the mechanical, thermal, and service properties of concretes, foams, and backfill materials. Evidence indicates that performance depends more on chemical complementarity than on maximizing waste content. Environmental benefits must also account for contaminant immobilization, carbon dioxide binding, and the energy and reagent demand of pretreatment. Mineralogy-informed machine-learning models may support inverse mixture design, but their transferability remains constrained by feedstock heterogeneity and limited standardized data. The framework links mineralogy, processing, reaction products, performance, and scale-up. Full article