Search Results (1,336)

The complexity of voltage stability assessment in modern smart grids has increased significantly with the growing penetration of renewable energy sources and the dynamic nature of load variations. Although standard analytical methods are accurate, they are computationally expensive and unsuitable for real-time applications. This paper proposes a hybrid analytical–machine learning framework for efficient voltage stability assessment and classification. The proposed approach consists of two stages. First, a power flow analysis is performed to compute the Fast Voltage Stability Index (FVSI) and quantify the proximity of the system operating conditions to voltage instability. Then, the FVSI values are converted into binary stability labels to formulate a supervised classification problem. In the second stage, the Extreme Gradient Boosting (XGBoost) algorithm is employed to learn the relationship between system operating variables and the corresponding stability states. The performance of the proposed method is evaluated on the IEEE 30-bus system and compared with that of conventional machine learning and deep learning models, such as Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Deep Neural Networks (DNNs). The simulation results show that the XGBoost-based framework outperforms the benchmark models in terms of classification accuracy, robustness, and computational efficiency. The proposed method provides a fast, reliable, and interpretable solution for real-time voltage stability monitoring. Therefore, it is suitable for modern smart grid applications. Full article

Most existing driver behavior classification technologies have focused on highways and other primary road infrastructures, despite secondary roads accounting for a disproportionately large number of traffic fatalities worldwide. Compared with highways, secondary roads present greater variability in road geometry, infrastructure quality, and traffic interactions, making driver behavior recognition considerably more challenging. This paper investigates the classification of driver behavior on secondary roads using machine learning techniques. Naturalistic driving data obtained from the publicly available UAH-DriveSet dataset were analyzed using two complementary feature groups describing lane detection and traffic status. Four supervised machine learning algorithms, namely, Logistic Regression (LR), gradient boosting (GB), Random Forest (RF), and Artificial Neural Networks (ANNs), were evaluated to classify driving behavior into three categories: Normal, Aggressive, and Drowsy. The extracted features were first analyzed through statistical profiling and exploratory feature analysis before training and evaluating the classification models. The experimental results show that gradient boosting consistently achieved the highest performance for both feature groups, attaining an overall classification accuracy of approximately 67% while providing balanced precision, recall, and F1-scores across all behavioral classes. Logistic regression and random forest produced competitive but lower performance, whereas the Artificial Neural Network yielded the lowest classification accuracy. The obtained results demonstrate the effectiveness of ensemble learning methods for driver behavior recognition under secondary-road conditions and highlight their potential for integration into intelligent driver monitoring and Advanced Driver Assistance Systems (ADASs). By enabling earlier identification of aggressive and drowsy driving behaviors on secondary roads, the proposed approach could support timely driver warnings and safety interventions, potentially reducing accident risk. Furthermore, the findings provide a benchmark for future machine learning models designed for real-world secondary-road environments, where driving conditions are more variable and challenging than on highways. Full article

Optical coherence tomography (OCT) is a high-resolution and non-contact optical imaging and sensing modality that provides depth-resolved cross-sectional visualization of retinal microstructures. It plays an important role in the assessment of retinal diseases, including age-related macular degeneration (AMD) and diabetic macular edema (DME). However, automated OCT image classification commonly relies on fully supervised models that require large-scale expert annotations, which are costly and time-consuming because of the complex layered anatomy and subtle pathological patterns present in retinal OCT images. To reduce annotation dependence, this study proposes a self-supervised representation learning framework with local dimensionality regularization for retinal OCT image classification. The proposed method estimates the local intrinsic dimensionality of learned representations and incorporates it into an asymptotic Fisher-Rao regularization objective to mitigate local dimensional degeneration and preserve fine-grained structural information. Logarithmic scaling and geometric averaging are further introduced to reduce sensitivity to outliers and improve optimization stability. Experiments on three independent OCT datasets achieved classification accuracies of 94.35%, 92.48%, and 92.56%, respectively, demonstrating competitive performance compared with mainstream self-supervised methods. These results demonstrate that explicitly modeling local feature geometry can improve the discrimination of sensor-acquired OCT images while reducing reliance on manual annotations, providing an effective approach for intelligent analysis of biomedical optical imaging data. Full article

The application of Machine Learning (ML) to bridge Structural Health Monitoring (SHM) is constrained by a lack of diverse and labelled datasets. Obtaining high-quality training data from operational infrastructure is inherently difficult because critical assets are typically repaired immediately upon the detection of defects, preventing the collection of data describing diverse failure modes. To address this scarcity and enable the prototyping of robust algorithms, this study presents a framework for generating synthetic modal frequencies using a calibrated Finite Element (FE) model of the S101 bridge. Aleatory uncertainties and environmental variability are incorporated through the stochastic variation of material properties and thermal loads derived from a 20-year climate record. Analysis of the generated dataset revealed that simulated thermal loads induced frequency shifts that often exceeded those caused by minor structural damage, confirming the necessity of training on environmentally representative data. The primary contribution of this work is an open-access, FAIR-compliant (Findable, Accessible, Interoperable, Reusable) synthetic dataset, intended to serve as a standardised benchmark for the SHM research community under conditions of combined structural and environmental uncertainty. To demonstrate the utility of the generated data, the performance of a supervised multi-layer perceptron and an unsupervised k-means clustering algorithm are evaluated, with the supervised approach achieving a maximum classification accuracy of 1.00. However, the framework also reveals a fundamental modelling limitation: the linear FE approach failed to replicate the physical response under pier settlement, producing frequency shifts an order of magnitude below those observed experimentally. Full article

Background/Objectives: Retinal imaging has considerable potential for monitoring Alzheimer’s disease (AD) neurodegeneration, as retinal ganglion cell dendritic atrophy within the inner plexiform layer (IPL) is an early event. We tested whether quantitative optical coherence tomography (OCT) speckle texture analysis combined with supervised machine learning could discriminate AD-related IPL alterations without exogenous contrast agents in a mouse model. Methods: Retinal explants from triple-transgenic AD mice (n = 7, aged 12 months) and C57BL/6 controls (n = 3, aged 15 months) were imaged ex vivo using a custom 1040 nm spectral-domain OCT system. Five grey-level co-occurrence matrix (GLCM) features were extracted from IPL volumes of interest (VOIs) and classified using a linear support vector machine (SVM). Results: AD and control IPL textures formed two completely separable clusters in a two-dimensional feature space defined by contrast and entropy (0°), achieving 100% VOI-level classification accuracy (95% CI: 96.4–100%). However, given the small sample size, VOI-level rather than animal-level validation, lack of histological confirmation, non-interleaved image acquisition, and differences in age/strain between groups, these results represent exploratory dataset separability rather than a validated diagnostic test. Conclusions: These findings demonstrate the feasibility of the ligand-free, texture-based OCT discrimination of IPL alterations, indicating a strong underlying optical signal. Adequately powered, in vivo longitudinal studies with matched controls, interleaved acquisition, animal-level cross-validation, and histological validation are required before any clinical translation. Full article

Background: Bariatric surgery improves metabolic health, but long-term transcriptomic remodeling of white adipose tissue (WAT) after Roux-en-Y gastric bypass (RYGB) remains incompletely defined. This study aimed to characterize longitudinal WAT gene-expression patterns after RYGB and prioritize candidate signatures of post-surgical adaptation using a publicly available dataset. Methods: We analyzed subcutaneous WAT transcriptomic data from women with obesity who underwent RYGB, with samples collected before surgery and at 2 and 5 years after surgery. Differential expression analysis was integrated with pathway enrichment, supervised machine-learning-based feature prioritization and classification, and SHAP-based model interpretation. Results: Differential expression and machine-learning analyses showed clear separation between baseline and post-surgery transcriptomic states. Pathway-level findings indicated reduced inflammatory and immune-related signaling, particularly across pathways related to phagosome function, lysosomal activity, antigen presentation, and host-defense responses after surgery. Gene-level analyses additionally suggested extracellular-matrix and metabolic remodeling. Machine-learning models distinguished baseline from post-surgery samples, while SHAP analysis identified genes with the strongest contributions to model predictions. Importantly, several statistically prioritized genes also showed high SHAP attribution, demonstrating concordance between univariate statistical significance and multivariate predictive relevance. This convergence suggests that the models captured biologically meaningful surgery-associated signals rather than purely data-driven classification artifacts. Conclusions: This study advances the interpretation of longitudinal adipose-tissue transcriptomic remodeling after RYGB by combining differential expression, pathway enrichment, supervised machine learning, and explainable AI within a unified framework. The integrated workflow prioritized candidate long-term remodeling genes, particularly immune/inflammatory and extracellular-matrix-related transcriptomic signatures, that warrant validation in independent cohorts. Full article

Land use and land cover (LULC) dynamics are critical indicators of environmental transformation and anthropogenic pressure on regional landscapes. Mirzapur, located in the transitional zone between the Indo-Gangetic Plain and the Vindhyan uplands in Northern India, represents a region characterized by ecological sensitivity, mineral-based industries, agricultural dependency, and rapid infrastructural growth. In recent decades, Northern India has experienced accelerated urbanization, population pressure, land fragmentation, and environmental stress, thus making systematic LULC monitoring crucial for sustainable resource management and policy planning. The present study examines the spatio-temporal changes in land use and land cover in Mirzapur for the years 2004, 2014, and 2024. The study employed a cloud-based platform and the Random Forest algorithm for supervised classification of multi-temporal satellite imagery. LULC maps were generated and post classification comparison was used to assess changes across the selected years. Accuracy assessment was conducted using standard validation metrics, including the Kappa coefficient, to evaluate classification. From 2004 to 2024, urban areas expanded by a relative increase of 169.36%, largely through the conversion of cropland, although the overall cropland area showed a slight increase due to agricultural expansion in other parts of the study area. A slight increase in forest cover was also observed during this period. Water bodies and barren lands declined, indicating ecological stress in the region. These changes reflect rapid urbanization, demographic pressure, and evolving socio-economic activities within the district. The LULC classification achieved overall accuracies of 96.50% (2004), 97.52% (2014), and 96.08% (2024), showing the reliability of the generated maps. The study demonstrates the effectiveness of cloud-based geospatial analysis combined with a machine learning algorithms for long-term LULC monitoring and provides valuable insights for sustainable land management and regional planning. Full article

Smart-diaper signals collected during routine care are affected by sensor noise, transmission gaps, variable event durations, and limited labels, making normal voiding (NV) and urinary incontinence (UI) difficult to distinguish using threshold-based detection alone. We developed a label-efficient time-series classification framework based on Context-Aware Temporal Contrastive Coding (CA-TCC) using the resistance ($R_{\mathrm{VAL}}$) channel of a smart-diaper sensor. Recordings from 97 older residents across three long-term care facilities were quality-filtered, aggregated at 3 min intervals, screened for candidate events, and interpolated to fixed-length inputs. CA-TCC was pre-trained on an unlabeled candidate-event pool and adapted using 4877 manually labeled events. The linear-probe, full fine-tuning, and class-aware pseudo-label retraining configurations were evaluated using participant-grouped five-fold cross-validation. The selected semi-supervised configuration achieved $82.12\pm 2.33$% accuracy, $82.12\pm 2.32$% macro-F1, and an AUC of $0.895\pm 0.018$ (mean ± 95% confidence interval), exceeding the strongest classical baseline by 4.80 macro-F1 percentage points. Its macro-F1 increased from $79.22\pm 1.84$% with 1000 labeled events to $81.97\pm 1.82$% with the full labeled set, whereas full fine-tuning showed greater fold-to-fold variability. Aggregated LIME analysis over 300 held-out events did not support localization of the model’s evidence to the event onset. These results indicate that contrastive pre-training can support smart-diaper voiding-pattern classification when labeled data are limited. Full article

Large language models are increasingly embedded in e-commerce assistants that recommend links to shopping destinations. When the candidate set contains adversarially crafted phishing URLs, a fluent model can recommend a malicious link with the same confidence it recommends a legitimate one, turning a helpful assistant into a delivery channel for fraud. Rather than treating phishing defense as per-link URL classification, this work frames the problem as recommendation-level exposure minimization: the output set itself must be secured, and a recommendation is counted as useful and safe only when it surfaces a benign destination and exposes no phishing URL. We first organize phishing URL constructions into four families spanning brand padding, typosquatting, homograph substitution, and subdomain impersonation, and use them to build candidate pools that mix benign links with plausible distractors. Evaluating four widely used open models under prompt-only defenses reveals a persistent gap: the strongest prompt baseline reaches only a 47.4 percent four-model average useful-safe rate, and weaker models fall below 20 percent even after careful prompting. We then present SAFER, a Security-Adaptive Filtering framework for Exposure-Minimized URL Recommendation. SAFER separates contextual selection from safety admission by coupling a deterministic lexical and structural pre-filter, an evidence-augmented single reasoning pass, and a deny-by-default post-verification stage with a deterministic fallback. The same deterministic URL evidence is injected before generation to condition the model’s reasoning and reused after generation to constrain which model-selected URLs may reach the user. SAFER issues exactly one model call per query, matching the prompt baselines, so its gains come from structure rather than additional inference. Across the four models SAFER raises the average useful-safe rate to 87.0 percent, a 39.6-point improvement over the best prompt baseline, and the deny-by-default stage yields a positive net gain for every model, largest where the model is weakest. Shrinking the deterministic layer’s brand coverage in a held-out analysis degrades the pipeline gracefully rather than collapsing it, indicating that within the range we tested its robustness does not rest solely on memorizing a fixed brand list. Additional robustness experiments confirm that SAFER transfers to real phishing URLs from the OpenPhish feed, generalizes to unseen attack families, maintains zero exposure on hard benign negatives, outperforms supervised URL classifiers as an admission gate, and maintains exposure minimization under realistic pool structures within the evaluated threat model. Full article

This article presents a systematic, evidence-weighted review of Deep Learning (DL) in farming, with a primary emphasis on the agricultural production stage and on four operational domains: precision crop management, precision livestock farming, soil and water resource management, and autonomous agricultural systems. Following a PRISMA 2020-oriented protocol, 56 primary studies (42 with quantitative results) were retained from an initial pool of 18,731 records. The reviewed literature reports applications in plant disease detection, weed recognition, yield prediction, fruit detection, livestock identification and health monitoring, soil-property estimation, crop-water-stress assessment, and robotic perception. High performance on controlled datasets, however, is frequently reported without external, temporal, or cross-site validation, making practical generalisation difficult to establish: in one widely cited benchmark, disease-classification accuracy fell from above 99% on held-out laboratory images to 31.4% on field-acquired images of the same classes. Persistent weaknesses include the limited availability of public benchmarks, inconsistent validation protocols, limited interpretability, fragmented data governance, and insufficient techno-economic analysis. The review argues that the next stage of agricultural AI should be judged less by isolated benchmark scores and more by field realism, reproducibility, deployment maturity, and practical usefulness. Unlike broad surveys that mainly catalogue architectures and applications, this review interprets the literature according to dataset representativeness, validation protocols, benchmarking transparency, deployment realism, and reproducibility, distinguishing algorithmic performance under controlled conditions from practical readiness for real farming environments. The most promising research directions include self-supervised and multimodal learning, explainable and privacy-preserving AI, edge-aware deployment, and hybrid process-informed models. Full article

Passive distributed acoustic sensing (DAS) is an attractive tool for monitoring geological CO₂ storage, but its dense, continuous recordings create a data-volume problem: a multi-month, multi-well archive yields enormous numbers of detector triggers, of which confirmed seismic events form a vanishingly small fraction, and conventional supervised classification is ill-posed when labels remain scarce and the negative class undefined because the non-event population is open-ended and spans noise families that vary over time and between wells. We present Argus, a sparse-label machine-learning workflow that converts continuous DAS recordings into a reproducible, auditable catalogue of event candidates. A deterministic front end reduces the archive to comparable trigger objects, each described by a 67-feature interpretable representation of its two-dimensional time–channel character (e.g., duration and channel span, detector-mask morphology, apparent moveout, inter-channel waveform coherence, and spectral shape); a retrieval-first machine-learning layer then ranks these triggers by their similarity, in this interpretable feature space, to a small seed catalogue of independently confirmed events, within an iterative human-in-the-loop process that introduces local supervised noise-rejection gates only for recurrent artefact families once they have been labelled. Applied to the CO2CRC Otway Stage 4 dataset—120 days of recordings on two wells, comprising roughly 23 TB and 14.14 million raw triggers—the workflow expanded a 39-event seed catalogue into 631 analyst-reviewed events, demonstrating complementarity with an independent template-matching analysis: two additional induced-event candidates were recovered, one within the CRC4 template-matching coverage and one on CRC7 during a CRC4 data gap. The induced-event class itself grew only from four to six candidates, and its counts are reported as a reviewed lower bound rather than a complete census. The result is a provenance-preserving, conservatively interpreted event inventory rather than an opaque classifier output, an outcome aligned with the reproducibility and audit requirements of CO₂ storage assurance. Full article

Reliable scene understanding under open-world conditions requires intelligent perception systems to accurately recognize known semantic categories while remaining robust to out-of-distribution (OOD) inputs, distribution shifts, and uncertain environmental conditions. This challenge becomes increasingly important for resource-constrained edge intelligence, where lightweight models are expected to provide reliable predictions without sacrificing computational efficiency. Although knowledge distillation has achieved remarkable success in compressing deep neural networks, existing methods primarily transfer classification semantics and often neglect the uncertainty representations that are critical for reliable open-set perception. To address this issue, we propose Residual Feature-driven Knowledge Distillation (RFKD), a lightweight uncertainty-aware distillation framework for reliable open-set scene understanding under distribution shift. Instead of directly distilling output confidence or energy scores, RFKD reconstructs uncertainty within the student’s latent feature space through a compact residual uncertainty branch. The proposed framework combines confidence-aware supervision, relational uncertainty distillation, and energy-guided relative ordering to preserve teacher-induced uncertainty geometry while enabling the student to learn discriminative feature-level uncertainty representations. The present study is evaluated on unimodal image data and does not claim empirical validation for multimodal perception. Extensive experiments on CIFAR-100 using multiple OOD benchmarks demonstrate that RFKD consistently improves uncertainty estimation while maintaining high computational efficiency. Compared with the ResNet-50 Teacher (Energy), RFKD increases the average AUROC from 0.7793 to 0.8446 while reducing the model size from 23.71 M to 11.29 M parameters and computational complexity from 1.31 G to 0.56 G FLOPs; the ImageNet-style student baseline obtains an AUROC of 0.7528. A score-specific sensitivity analysis shows that the energy detector is strongest when the auxiliary ordering score uses the residual branch alone, whereas moderate coupling with the student’s log-sum-exp potential improves the standalone OOD head. These results demonstrate that explicitly modeling representation-level uncertainty offers an effective and efficient solution for reliable scene understanding, providing a practical reliability enhancement for future intelligent perception systems operating in open and dynamic environments. Full article

Black-sand mining is an under-studied anthropogenic pressure on arid coastal environments, where sparse vegetation and slow natural recovery limit conventional impact assessment. Despite the recent expansion of heavy-mineral extraction along the Mauritanian coast, no spatio-temporal analysis has quantified its effects on land cover and surface-water dynamics in the N’Diago region. We analysed the N’Diago–LGUWYCHICH coastal sector (south-western Mauritania) using three Landsat scenes (2014, 2020, and 2026) in QGIS (free and open-source Geographic Information System), four spectral indices (NDVI-Normalized Difference Vegetation Index, NDWI-Normalized Difference Water Index, BSI- Bare Soil Index, and CI -Coloration Index), supervised Support Vector Machine classification and a 30 m SRTM Digital Elevation Model over a 97.82 ha area of interest. Bare soil dominated the landscape at every date (92.6–95.7%) and vegetation stayed below 7%, indicating that canopy-based metrics underestimate disturbance in this setting. The clearest change was hydrological: an inland water body shrank from 1.02 ha in 2020 to 0.40 ha in 2026, a 60.6% loss. This individual inland water body, measured directly and independently from the NDWI index, is our primary hydrological observation: the wet feature was smaller in the 2026 image than in the 2020 image and was located near the mapped concession areas, but the available data do not establish the cause of this change. Similar bare-soil and colour index values (BSI ≈ 0.23, CI ≈ 0.79–0.80) were observed in the processed images, while residual seasonal and radiometric differences between the Landsat 8 and DOS-corrected Landsat 9 products cannot be excluded; BSI and CI are treated only as candidate or contextual spectral patterns, so any delineation of the disturbed footprint is provisional and requires confirmation from independent field data. This study illustrates a low-cost exploratory workflow that may support preliminary monitoring in data-scarce arid coastal settings, pending validation with denser time series and field observations. Full article

Automatic detection of stress in social media text holds promise for supporting digital mental health, but most existing Transformer-based approaches are opaque and computationally demanding. This work presents DR-Transformer, a Dual-Regularized Transformer that combines two complementary mechanisms: (i) a group sparsity penalty ($L_{2,1}/L_2$ elastic net) applied to the query and key projection matrices of every attention head, which encourages whole-row sparsity, producing more concentrated and inspectable attention patterns; (ii) a supervised contrastive loss on the [CLS] projection, which organizes the latent space according to the stress label. The architecture is intentionally lightweight (six layers, eight heads, 256-dim embeddings; ∼9.5 M parameters) and runs entirely on consumer-grade hardware (NVIDIA GTX 1660, 6 GB). Experiments on the publicly available Dreaddit dataset (binary stress classification, 2838 train/715 test segments) compare DR-Transformer against Logistic Regression, BiLSTM, a Standard Transformer of identical architecture, and MentalBERT. Across five seeded runs, DR-Transformer (Full) reaches $F_1=0.876$ (bootstrap 95% CI $0.852$–$0.898$), outperforming the Standard Transformer ($F_1=0.842$; McNemar $p<0.001$ with Bonferroni correction) and performing comparably to the much larger MentalBERT ($F_1=0.879$; $p=0.421$). Sparse regularization increases the fraction of near-zero attention weights (below $0.01$) from $0.215$ to $0.682$, while the supervised contrastive loss improves the silhouette score of [CLS] embeddings from $0.312$ to $0.483$. Dual regularization thus combines accuracy, efficiency, and structurally induced attention concentration in a single model which can be trained without specialized infrastructure. We use the term “interpretable” throughout in this restricted, structural sense—to refer to concentrated and inspectable attention—rather than in the sense of established causal or mechanistic faithfulness; this is only partially and indirectly supported by our token deletion analysis. Full article

In live-streaming e-commerce systems, massive volumes of user-generated danmaku constitute a critical yet underutilized source of business intelligence. However, transforming such unstructured, noisy, and highly context-dependent textual data into structured and actionable knowledge remains a fundamental challenge for enterprise information systems. To address this issue, this study proposes TaSC-LLM, an LLM-enabled topic recognition method for constructing interpretable topic measurements from unstructured user-generated content. The proposed framework integrates topic taxonomy construction and zero-shot classification into a unified semantic reasoning pipeline. Unlike conventional topic modeling or supervised classification approaches, TaSC-LLM leverages chain-of-thought reasoning, multi-stage taxonomy induction, sliding window context modeling, and self-consistency verification to eliminate reliance on predefined label spaces and annotated training data. This design allows the system to dynamically construct and update topic taxonomies while ensuring interpretability, robustness, and cross-scenario adaptability. Empirical evaluation on three large-scale live-streaming e-commerce danmaku datasets shows that TaSC-LLM achieves strong taxonomy coverage, classification accuracy, and agreement with expert annotations. The findings suggest that LLM-based reasoning can help convert unstructured user-generated text into interpretable topic measures for downstream empirical and managerial analysis. While the present evaluation is conducted offline, TaSC-LLM provides a methodological foundation for future business applications that can be further examined under multi-session, multi-platform, and deployment-oriented conditions. Full article