Search Results (412)

Traditional intrusion detection systems for IoT networks achieve high classification accuracy but lack interpretability and actionable incident-response capabilities, limiting their operational value in security-critical environments. This paper presents a graph-based multi-agent framework that integrates ensemble machine learning with Large Language Model (LLM)-powered incident report generation via Retrieval-Augmented Generation (RAG). The system employs a three-phase architecture: (1) a lightweight Random Forest binary pre-detection, achieving 99.49% accuracy with a 6 MB model size for edge deployment; (2) ensemble classification combining Multi-Layer Perceptron, Random Forest, and XGBoost with soft voting and SHAP-based feature attribution for explainability; and (3) a ReAct-based summary agent that synthesizes classification results with external threat intelligence from Web search and scholarly databases to generate evidence-grounded incident reports. To address the challenge of evaluating non-deterministic LLM outputs, we introduce custom RAG evaluation metrics—faithfulness and groundedness implemented via the LLM-as-Judge framework. Experimental validation on the ACI IoT Network Dataset 2023 demonstrates ensemble accuracy exceeding 99.8% across 11 attack classes; perfect groundedness scores (1.0), indicating all generated claims derive from the retrieved context; and moderate faithfulness (0.64), reflecting appropriate analytical synthesis. The ensemble approach mitigates individual model weaknesses, improving the UDP Flood F1 score from 48% (MLP alone) to 95% through soft voting. This work bridges the gap between high-accuracy detection and trustworthy, actionable security analysis for automated incident-response systems. Full article

Real-time facial emotion recognition remains challenging due to the high dimensionality and computational cost of dense facial representations, which limit their applicability in resource-constrained and real-time scenarios. This study proposes a compact, anatomically informed angular facial representation for efficient, interpretable emotion recognition under real-time constraints. Facial landmarks are first extracted using a standard landmark detection framework, from which a reduced facial mesh of 27 anatomically selected points is defined. Internal geometric angles computed from this mesh are analyzed using temporal variability and redundancy criteria, resulting in a minimal set of eight angular descriptors that capture the most expressive facial dynamics while preserving geometric invariance and computational efficiency. The proposed representation is evaluated using multiple supervised machine learning classifiers under two complementary validation strategies: stratified frame-level cross-validation and strict Leave-One-Subject-Out evaluation. Under mixed-subject stratified validation, the best-performing model (MLP) achieved macro-averaged F1-scores exceeding 0.95 and near-unity ROC–AUC values. However, subject-independent evaluation revealed reduced generalization performance (average accuracy ≈55%), highlighting the influence of inter-subject morphological variability embedded in absolute angular descriptors. These findings indicate that a minimal angular geometric encoding provides strong intra-subject discriminative capability while transparently characterizing its cross-subject generalization limits, offering a practical and interpretable alternative for data- and resource-constrained real-time scenarios. Full article

Accurate trajectory tracking of redundant manipulators is difficult because the controller must simultaneously model local couplings between adjacent joints and global dependencies across the whole kinematic chain. Existing reinforcement learning methods typically employ multilayer perceptrons, which do not explicitly exploit manipulator structure and therefore show limited stability and representation ability in high-dimensional continuous control tasks. This paper proposes GT-TD3, a Graph Transformer-enhanced-Twin Delayed Deep Deterministic Policy Gradient framework, for redundant manipulator trajectory tracking. The proposed actor first converts the raw system state into joint-level node features and uses a graph neural network to extract local kinematic coupling information. A Transformer is then employed to capture long-range dependencies among joints. To strengthen the use of structural priors, topology- and distance-related bias terms are incorporated into the attention mechanism, enabling the network to encode manipulator structure during global feature learning. Experiments on a 7-DoF KUKA iiwa manipulator in PyBullet demonstrate that GT-TD3 outperforms MLP, pure GNN, and pure Transformer baselines in tracking performance. The proposed method achieves more stable training, faster convergence, and smoother and more accurate end-effector motion. The results show that the integration of local graph modeling and structure-aware global attention provides an effective solution for high-precision trajectory tracking of redundant manipulators. Full article

Accurately predicting the peak height of the F2 layer (hmF₂) is crucial for radio communications, satellite navigation, and space weather studies, yet traditional empirical models often lack precision. To address this, we developed a global hmF₂ prediction model using a Multilayer Perceptron (MLP) neural network, based on COSMIC radio occultation observations from 2007 to 2023. Evaluated on independent test sets from 2014 and 2019, the MLP model achieved correlation coefficients of 0.877 and 0.853 with root mean square errors of 22.3 km and 19.1 km, respectively, significantly outperforming the machine learning approaches like XGBoost and Transformer. The model also demonstrated strong generalization on an independent validation set constructed from 2014 and 2019 GIRO data, with a correlation of 0.785 and RMSE of 26.1 km, surpassing NPHM, and XGBoost. SHAP interpretability analysis identified geographic latitude, cosine of latitude, solar F10.7 index, and annual/daily harmonic terms as the most influential physical features. Error analysis showed a mean prediction error of −3.2 km and a standard deviation of 21.8 km, with stable performance during quiet periods and larger errors primarily during disturbed conditions. This study provides a reliable tool for high-accuracy hmF₂ prediction and enhances the understanding of the physical mechanisms controlling its variability. Full article

A neural network simulation–regression method was developed to assess the impact of sanctions on the level of administrative offences under fragmented, noisy, and short administrative time series. The study addresses the problem of quantifying and predicting changes at the offence level as a sanction size function, using detection probability, prior violation level, compliance costs, and auxiliary contextual factors. The proposed framework combines a hybrid MLP–LSTM neural network, double machine learning-based orthogonal causal estimation, the simulation-based generation of counterfactual scenarios through domain randomization, multiple imputation for missing data, debiasing procedures, and ensemble uncertainty estimation. The contribution to administrative law consists of a quantitative tool creation for substantiating and optimising sanction policy, assessing heterogeneous effects, and supporting evidence-based rulemaking and law enforcement decisions. In comparative experiments, the developed method achieved an RMSE of 8…12%, a prediction accuracy of 93…96%, an overall accuracy of 95%, a precision of 94%, a recall of 93%, and an F1-score of 93.5%, thereby outperforming contemporary econometric, simulation, causal machine learning, and predictive machine learning approaches used for sanction effect modelling. Additional verification through nonparametric statistical testing confirmed that the proposed model’s superiority over the compared algorithms is statistically significant across the main quality metrics, which strengthens the evidence for its robustness and practical value in sanction policy analysis under fragmented administrative data conditions. Full article

Hypertension is a leading global health risk and requires accurate and continuous monitoring for effective management. Although photoplethysmography (PPG) is a promising non-invasive modality for cuffless blood pressure (BP) assessment, many existing approaches (especially raw-signal deep learning) are vulnerable to data leakage, overfitting on small datasets, limited interpretability, and poor performance on minority BP stages. To address these limitations, we propose a robust and physiologically grounded framework for multi-class BP stage classification based on interpretable PPG features. Our approach centers on a comprehensive multi-domain feature engineering pipeline that extracts 124 PPG features, including demographic, morphological, functional decomposition, spectral, nonlinear dynamics, and clinical composite indices. We apply rigorous preprocessing and feature selection prior to model training. We validate the framework on two datasets: PPG-BP dataset (657 segments, 4 classes) for benchmarking and PulseDB (283,773 segments, 3 classes) to assess scalability. We evaluate the proposed framework using a segment-level train/test split, appropriate for assessing intra-subject BP tracking after initial personalization. For the PulseDB dataset, this follows the protocol established by the dataset creators, while for the PPG-BP dataset, it enables direct comparison with prior work given practical dataset constraints. On PPG-BP, LightGBM trained on the selected features achieved macro-F1 = 0.78 and accuracy = 0.74, outperforming comparable deep-learning models. On the PulseDB, a custom Residual MLP achieved accuracy = 0.81 and macro-F1 = 0.79, supporting generalization at scale. These results show that the proposed feature-based approach can outperform complex end-to-end deep-learning models on small datasets while providing improved interpretability. This work establishes a reliable and transparent pathway toward clinically viable continuous BP staging, moving beyond black-box models toward physiologically grounded decision support. Ablation analysis reveals that engineered features provide most of the predictive power (F1 = 0.911), while raw PPG features alone achieve modest performance (F1 = 0.384). For the minority hypertension stage 2 (HT-2) class, a bootstrap 95% confidence interval of [0.762, 1.000] is reported, reflecting uncertainty due to limited sample size. Full article

Unbalanced supply voltage (USV) represents a critical power quality challenge in industrial environments, significantly degrading the performance, efficiency, and operational lifespan of three-phase induction motors. Accurate real-time estimation of sequence impedances ($Z_a,Z_b,Z_c$) and detection of the Negative Voltage Factor (NVF) are essential for effective condition monitoring and preventive maintenance strategies. While existing machine learning methods have demonstrated promising accuracy, they often rely on manual feature engineering, lack hierarchical representation learning, and treat impedance estimation and fault detection as isolated tasks. This paper proposes a unified Deep Multi-Task Learning framework that leverages Residual Multilayer Perceptron (ResMLP) architectures for feature-based learning and Temporal Convolutional Networks (TCNs) for end-to-end raw signal learning. Our contributions include: (1) introduction of a Multi-Head ResMLP architecture that jointly optimizes phase impedance and fault detection, achieving superior NVF accuracy (MAE = 0.0007) and a fault detection F1-score of 0.8831; (2) investigation of raw-voltage TCN models for voltage-only diagnostics, with analysis of the trade-offs between end-to-end learning and feature-based approaches; (3) extensive ablation studies demonstrating the impact of network depth, data augmentation, and training protocols on model generalization; and (4) deployment of PyTorch (v2.0.1)-based models suitable for embedded systems with real-time inference capabilities (2.3 ms per prediction). Experimental validation on a 1.1 kW three-phase motor dataset under diverse load conditions (0–10 Nm) and USV magnitudes (5–15 V) confirms the robustness and practical applicability of the proposed approach for industrial fault diagnosis and condition monitoring systems. Full article

Non-Intrusive Load Monitoring (NILM) systems are increasingly applied in residential and commercial environments to disaggregate energy consumption without requiring additional hardware sensors. The integration of Machine Learning (ML) techniques has enhanced the accuracy and efficiency of load identification and classification in smart meter-based systems. This study presents a systematic review and meta-analysis aimed at identifying, classifying, and quantitatively evaluating ML models applied to NILM. Searches were conducted in the IEEE Xplore and Scopus databases, restricted to peer-reviewed publications from 2017 to 2024. Thirty studies met the eligibility criteria and were included in the quantitative synthesis using a random-effects meta-analysis model (DerSimonian–Laird estimator). The primary effect measure was the F1-score. Statistical analyses were performed using R (version 4.5.0) and Python (version 3.10.0), including heterogeneity assessment and subgroup analyses according to model type. Hybrid models, such as SVDT-KNN-MLP, LE-CRNN, and RBFNN-MOGA, achieved the highest pooled F1-scores, although supported by a limited number of studies. Traditional approaches, including CNN, KNN, and Random Forest, demonstrated consistently strong performance and broader validation, whereas Boosted Trees and RNN-based models showed lower or more variable results. Substantial heterogeneity was observed across studies, highlighting the need for dataset standardization, reproducible evaluation frameworks, and further validation of emerging hybrid architectures in diverse operational scenarios. This study contributes by providing a quantitative synthesis of machine learning models applied to NILM using a structured PRISMA-based methodology and subgroup analysis by model architecture. Unlike previous narrative reviews, this work integrates scientometric analysis with meta-analytic performance aggregation, offering a consolidated and comparative evidence base for future NILM research. Full article

Reliable seed accession identification underpins germplasm conservation, traceability and breeding; however, conventional assays remain destructive, labour-intensive and difficult to scale. Here, visible–near-infrared–shortwave infrared (VIS–NIR–SWIR) hyperspectral imaging (HSI; 449.54–2399.17 nm; 563 bands) was used to classify 32 grain–legume accessions (n = 3200 seeds; 100 seeds per accession), comprising 30 common bean (Phaseolus vulgaris L.) landraces plus two outgroup legumes (Vigna angularis (Willd.) Ohwi & Ohashi and Cajanus cajan (L.) Huth). Each seed was represented by one ROI-averaged spectrum obtained from mean representative pixels within a standardised 10 × 10 pixel window at the centre of each seed. A fixed stratified 70:30 seed-level training:test partition was used, with 70 seeds per accession (n = 2240) reserved for fully independent training and 30 seeds per accession (n = 960) reserved as a fully independent test set. Principal component analysis (PCA) captured 97.42% of the spectral variance in the first three components (PC1 = 63.34%, PC2 = 23.78%, and PC3 = 10.31%). One-versus-rest wavelength association mapping revealed a maximum R² of 0.775 at 461.37 nm, and ReliefF concentrated the strongest reduced-band signal within 449.54–456.30 nm and 577.02–597.54 nm. In the original ReliefF-selected 16-band benchmark, the subspace discriminant reached 68.25% macro-F1 and 68.54% balanced accuracy; after edge-band trimming, the alternative 16-band configuration decreased to 60.67% and 60.94%, respectively. With respect to the full-spectrum sensitivity benchmark, linear discriminant analysis achieved 96.35% balanced accuracy, followed by linear SVM (94.17%). Deep learning trained directly on the full 563-band spectra reached 84.90% test accuracy, 84.47% macro-F1, 86.27% precision and 84.90% recall, with MLP_Wide outperforming the convolutional, recurrent and attention-based alternatives. Overall, under controlled laboratory conditions, this benchmark shows that accession discrimination is driven mainly by visible-domain contrasts in the most compact representations, whereas the full spectral context remains important for the most confusable accessions and for cautious future sensor design. The reduced-band findings should therefore be interpreted as exploratory guidance for sensor design rather than as a validated deployment-ready specification. Full article

Fault diagnosis in modern power systems increasingly depends on unstructured operation and maintenance (O&M) logs, yet real-world logs are often small in scale and highly imbalanced across fault types, which degrades the generalizability of standard neural models. This paper proposes a graph-augmented diagnostic framework that integrates imbalance-aware knowledge extraction with interpretable reasoning. The framework consists of three stages: (1) domain adaptation of a BERT–BiLSTM–CRF NER model and a BERT–MLP RE model using an imbalance-aware training recipe that combines Low-Rank Adaptation (LoRA), a mixed focal–range loss, and undersampling; (2) construction of a power-system knowledge graph that organizes extracted entities and relations (e.g., fault devices, abnormal phenomena, causes, and handling measures); and (3) a graph-augmented assistant agent that reuses the NER model as a graph-aware retriever within a retrieval-augmented generation (RAG) architecture to support contextualized and interpretable diagnostic reasoning. Experiments on 3921 real-world fault-processing logs show consistent gains: NER reaches 92.0% accuracy and 71.3% Macro-F1 (vs. 80.3% and 63.2%), and RE achieves 88.0% accuracy and 70.1% F1 (vs. 82.1% and 60.4%), while reducing average training time per epoch by about 18%. These results demonstrate an efficient and practical path toward robust log-based fault diagnosis under scarce and imbalanced data. Full article

This paper investigated the feasibility of snapshot multispectral fluorescence imaging for nondestructive identification of cold stress in pepper plants. Fluorescence spectra were obtained by exciting the plant with a 405 nm ultraviolet LED. The plants were grown under three temperature conditions: 17 °C (control), 10 °C (moderate cold stress), and 5 °C (severe cold stress). Raw fluorescence spectra extracted from the demosaiced snapshot images were used as inputs for a deep-learning pipeline consisting of feature extraction, an encoder–decoder GRU, and a multilayer perceptron (MLP), and the results were compared with conventional machine learning classifiers, including linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and a Gaussian support vector machine (G-SVM). Tukey’s HSD test indicated that the proposed deep-learning model achieved the highest cross-validation accuracy and consistently produced superior classification metrics (accuracy of 85.7%, precision of 85.3%, recall of 85.3%, F1-score of 85.2). The trained model was further applied to hyperspectral cubes to generate classification maps; however, moderate misclassification was observed, consistent with the overall prediction performance. Full article

Credit risk prediction is essential for financial institutions to effectively assess the likelihood of borrower defaults and manage associated risks. This study presents a comparative analysis of deep learning architectures and traditional machine learning models on imbalanced credit risk datasets. To address class imbalance, we employ three resampling techniques: Synthetic Minority Over-sampling Technique (SMOTE), Edited Nearest Neighbors (ENN), and the hybrid SMOTE-ENN. We evaluate the performance of various models, including multilayer perceptron (MLP), convolutional neural network (CNN), long short-term memory (LSTM), gated recurrent unit (GRU), logistic regression, decision tree, support vector machine (SVM), random forest, adaptive boosting, and extreme gradient boosting. The analysis reveals that SMOTE-ENN combined with MLP achieves the highest F1-score of 0.928 (accuracy 95.4%) on the German dataset, while SMOTE-ENN with random forest attains the best F1-score of 0.789 (accuracy 82.1%) on the Taiwanese dataset. SHapley Additive exPlanations (SHAP) are employed to enhance model interpretability, identifying key drivers of credit default. These findings provide actionable guidance for developing transparent, high-performing, and robust credit risk assessment systems. Full article

Rapid Internet of Things (IoT) expansion creates security risks due to resource limits and evolving botnets. While Graph Neural Networks (GNNs) offer accuracy, their computational demands hinder real-time edge deployment. This study presents IoTGuard, based on a ‘Hetero-MLP’ architecture. The model replaces costly message passing with 8-dimensional categorical embeddings to capture protocol semantics. To avoid topology overfitting, L3 identifiers were excluded, relying on 13 L4 attributes selected via Pearson correlation. Evaluations on the NF-BoT-IoT-v2 dataset (37.7 M samples) demonstrate a 12.17 KB (INT8) footprint via post-training quantization. This represents a 1.9× size reduction, enabling independent operation on ARM Cortex-M7 platforms (Arm Ltd., Cambridge, UK) at 37,093 requests per second. The framework achieves a DDoS F1-score of 0.9943 with a false-positive rate of 0.0054. Comparative analysis confirms that while Random Forest is accurate, Hetero-MLP reduces parameters by 25.4× versus standard GAT models. The proposed approach balances detection depth with edge constraints, offering scalable critical infrastructure protection. Full article

Regional epidemic risk prediction requires holistic modeling of heterogeneous data sources such as demographic structure, health capacity, geographical features and human mobility. In this study, a unique and multi-modal epidemiological data set integrating demographic, health, geographic and mobility indicators of Türkiye and its neighboring countries was collected. Türkiye’s neighboring countries are Greece, Bulgaria, Georgia, Armenia, Iran, and Iraq. This dataset, created by combining raw data from these neighboring countries, provides a comprehensive regional representation that allows for both quantitative classification and spatial mapping of epidemiological risk. To address the class imbalance problem, Conditional GAN (CGAN), a class-conditional synthetic example generation approach that enhances high-risk category representation was used. In this study, we proposed a multi-view deep learning model named MV-RiskNet, which effectively models the multi-dimensional data structure by processing each view into independent subnetworks and integrating the representations with an attention-based fusion mechanism for regional epidemic risk prediction. Experimental studies were compared using Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Autoencoder classifier, and Graph Convolutional Network (GCN) models. The proposed MV-RiskNet with CGAN model achieved better results compared to other models, with 97.22% accuracy and 97.40% F1-score. The generated risk maps reveal regional clustering patterns in a spatially consistent manner, while attention analyses show that demographic and geographic features are the dominant determinants, while mobility plays a complementary role, especially in high-risk regions. Full article

Partial discharge (PD) signals in gas-insulated switchgear (GIS) exhibit complex characteristics, and single-modal feature recognition methods face limitations in achieving satisfactory diagnostic accuracy due to incomplete fault information representation. This paper proposes a multi-modal fault diagnosis framework that effectively integrates complementary information from different sensing modalities to improve defect identification performance. First, PRPD time-domain statistical features from HFCT measurements and frequency-domain features from UHF signals are extracted to construct a comprehensive hybrid feature set. Z-score normalization is applied to eliminate scale differences between heterogeneous features. Principal component analysis (PCA) is then employed for dimensionality reduction, preserving essential discriminative information while removing redundancy. Finally, a ResNet-MLP classifier with skip connections is designed to enhance nonlinear feature extraction and alleviate gradient vanishing problems in deep network training. Experimental validation on four typical defect types—protrusion defect, floating discharge, metal particle discharge, and surface discharge on insulator—demonstrates that the proposed method achieves 99.38% classification accuracy on the test set, with consistently high precision, recall, and F1-score across all categories. The proposed approach significantly outperforms standard MLP without residual connections, achieving 98.94% ± 0.49% accuracy compared to 95.47% ± 3.72% over 20 independent runs, demonstrating superior diagnostic accuracy and generalization capability for GIS insulation fault diagnosis. Full article