Search Results (1,705)

High-dimensional environmental covariates are increasingly available for digital soil mapping (DSM), but their effective use depends on both the feature-selection strategy and the predictive model architecture. However, systematic evidence remains limited regarding how different metaheuristic feature-selection methods interact with standalone and hybrid learning models across multiple soil and groundwater prediction tasks. This study systematically evaluated the interactions between 10 metaheuristic feature-selection algorithms and 13 predictive models, including random forest (RF), convolutional neural network (CNN), recurrent architectures, CNN–recurrent neural network (RNN) hybrids, squeeze-and-excitation (SE)-enhanced hybrids, and iTransformer-based hybrids, across four prediction tasks involving soil organic carbon (SOC), soil–water extract electrical conductivity (ECe), apparent electrical conductivity (ECa), and groundwater level (GWL) in Xinjiang, China. A total of 149 candidate environmental covariates were considered for ECe, SOC, and ECa, whereas 122 candidate covariates were considered for GWL. The results showed that no single feature-selection method consistently performed best across all four targets; instead, predictive performance depended on the interaction among the feature-selection strategy, predictive architecture, and target variable. CNN–RNN hybrid architectures generally achieved higher predictive performance than standalone models, although their benefits varied among prediction targets. The best-performing combinations yielded coefficient of determination (R²) values of 0.9826, 0.6981, 0.8429, and 0.8085 for GWL, SOC, ECe, and ECa, respectively. These findings indicate that target-specific compatibility, rather than aggressive dimensionality reduction or a universally superior algorithm, is a key determinant of predictive performance in high-dimensional DSM. By demonstrating that feature-selection effectiveness is jointly influenced by model architecture and target characteristics, this study provides a methodological reference for developing target-specific digital soil mapping models in arid regions. Full article

The Tianjin Container Freight Index (TCI) is a key benchmark for container shipping prices in northern China and is strongly driven by macroeconomic conditions and trade fluctuations. This study proposes a PCC–CNN–GRU hybrid deep learning framework for TCI forecasting using 25,870 daily observations from 13 April 2015, to 1 January 2024. The model combines Pearson Correlation Coefficient (PCC)-based feature selection, convolutional neural networks (CNN) for local temporal feature extraction, and gated recurrent units (GRU) for capturing long-term dependencies, thereby addressing the nonlinear and nonstationary characteristics of TCI data. Empirical results show that the proposed model achieves an R² of 91.24%, outperforming standalone CNN, GRU, and classical ARIMA and VAR models. The model demonstrates strong robustness to structural changes and noise, enhancing its suitability for complex market environments. The integrated framework provides reliable forecasting support for shipping companies, logistics planners, and policymakers in pricing, capacity planning, and sustainable maritime operations. This study contributes to the growing integration of intelligent forecasting methods with regional freight index analysis and supports the digital transformation of the container shipping industry. Full article

Accurate continuous prediction of lower-limb joint angles from surface electromyography (sEMG) remains challenging because of the nonlinear, non-stationary, and subject-specific nature of sEMG signals, which can reduce robustness and lead to degraded prediction accuracy during highly dynamic gait phases. In this study, we propose RTGNet, a dual-branch deep learning framework for lower-limb joint-angle prediction from multichannel sEMG. The method constructs two feature views from the same sEMG stream: recurrence-plot (RP)-based representations for nonlinear texture characterization and time-series sequences for long-term temporal dependency modeling. These views are processed by a convolutional neural network (CNN) with a convolutional block attention module (CBAM) and a bidirectional long short-term memory network (BiLSTM), respectively, and integrated through an adaptive gated fusion mechanism. An enhanced Huber-TopK loss is further employed to emphasize samples with large prediction errors. Experiments on the SIAT-LLMD dataset under an offline cross-subject evaluation setting show that RTGNet achieves a mean absolute error (MAE) of 3.87°, a root mean square error (RMSE) of 5.25°, and an $R^2$ of 0.81 during walking, as well as an MAE of 4.53°, an RMSE of 6.47°, and an $R^2$ of 0.84 during stair ascent. The proposed framework outperforms temporal-only and RP-based baselines, and ablation results further support the effectiveness of the gated fusion strategy and CBAM attention. Overall, these results suggest that integrating recurrence texture and temporal dynamics is a promising strategy for sEMG-driven joint-angle prediction and provides a useful basis for future exoskeleton control-oriented studies. Full article

Background/Objectives: Automatic sleep staging from polysomnography (PSG) is particularly challenging in clinical cohorts with neurological disorders. Methods: This study presents SleepStageNet, a compact model (1.075M parameters) that integrates a dual-branch convolutional epoch encoder, feature gating, a bidirectional gated recurrent unit, and multi-head self-attention for five-class staging from five PSG channels (C3, C4, EOG1, EOG2, and chin EMG). The individual operations are adapted from established architectures; the study contribution is their compact integration and controlled evaluation in an Indian acute stroke cohort. Results: Of the 100 recordings in the Indian Sleep Polysomnography (iSLEEPS) resource, 95 satisfied the five-channel extraction criteria, yielding 78,323 annotated epochs. Subject-independent 10-fold stratified group cross-validation produced an accuracy of 73.91 ± 2.24%, a macro F1-score of 67.29 ± 1.96%, and a Cohen’s $\kappa$ of $0.634\pm 0.029$ (sample standard deviations). A matched single-branch encoder obtained $\kappa =0.635$ (full minus single branch: $\Delta \kappa =-0.001$, Holm-adjusted $p=0.846$), while matched C4-only input obtained $\kappa =0.600$ (full minus C4-only: $\Delta \kappa =0.033$, Holm-adjusted $p=0.008$). Grad-CAM and temporal attention visualizations provided qualitative evidence of physiologically plausible focus, while channel occlusion quantified the contribution of each signal. Without fine-tuning, a 10-model ensemble obtained $\kappa =0.614$ on ISRUC-SLEEP Subgroup III (10 healthy subjects; 8889 epochs). Conclusions: These results establish a reproducible reference for this clinical cohort while identifying the need for broader external and prospective validation. Full article

Reliable digital twins of complex nonlinear systems require not only high predictive accuracy but also physical consistency, robustness under degraded operating conditions, and explicit uncertainty handling. Purely data-driven models often suffer from poor generalization, unphysical behaviors, and limited interpretability when facing noisy measurements and unseen operating conditions. This paper introduces a knowledge-guided physics-informed hybrid learning framework that integrates recurrent neural networks with Unscented Kalman Filter (UKF) state estimation and embedded thermodynamic constraints within a unified uncertainty-aware architecture. The proposed PI-LSTM-UKF framework achieves competitive predictive accuracy and improved physical consistency relative to the residual-learning hybrids by tightly integrating physics-informed recurrent learning, thermodynamic constraints, and sequential UKF state estimation. While the UKF provides robust recursive correction under noisy measurements during closed-loop operation, the physics-informed Long Short-Term Memory (PI-LSTM) learns nonlinear corrections and long-term dynamics that cannot be captured by the linear model alone. The proposed framework is systematically benchmarked against a hierarchy of seven modeling approaches, including Dynamic Mode Decomposition with control (DMDc), Sparse Identification of Nonlinear Dynamics (SINDy), and residual-learning variants based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). High-fidelity Simscape simulations of a Rankine-cycle steam turbine system are used as a challenging simulation-based case study. Results show that the knowledge-guided hybrid approach achieves competitive predictive accuracy, improved physical consistency, and robust performance under an unseen load profile, severe thermodynamic degradation, valve hysteresis, and substantially elevated sensor noise. The framework provides a promising simulation-based foundation for uncertainty-aware digital twins of nonlinear thermal power systems. Validation using operational plant data remains necessary before its application to real-time monitoring and predictive maintenance. Full article

Reliable navigation in GPS-denied environments remains a critical challenge for autonomous vehicles (AVs), particularly in complex indoor and urban settings. GPS-based localization systems often fail under these conditions, highlighting the need for resilient multimodal solutions. In this article, we present a radar-assisted tracking system that integrates LiDAR and inertial measurements within a sensor-fusion architecture to achieve robust navigation. The principal methodological contribution is a unified tracking and prediction framework that combines Bayesian state estimation with learning-based temporal prediction, enabling accurate tracking while continuously forecasting the slave robot’s short-term future state from mapping observations generated by the master robot, with a typical end-to-end perception-to-action latency of 20–60 ms. The communication and prediction forecasting module operates with an update interval below 35 ms, enabling real-time cooperative robotic operation. Sensor data are fused through a pipeline incorporating Gaussian Mixture Models (GMMs) for post-processing, which helps mitigate the limitations associated with individual sensors during edge processing. Moreover, Kalman filtering is employed to mitigate sensor noise and drift, thereby improving state estimation accuracy through trajectory smoothing. The fused spatiotemporal information is subsequently exploited by a Convolutional Recurrent Neural Network (CRNN) coupled with a Nonlinear Autoregressive model with eXogenous Inputs (NARX) to model the robot’s motion dynamics and provide short-horizon state prediction. Through simulations and real-world indoor experiments conducted in GPS-denied environments, we validate the system’s ability to provide accurate and continuous pose estimation with low localization errors. Experimental results show that the proposed framework achieves root-mean-square errors of 0.12 m, 0.15 m, and 0.28 m along the X, Y, and Z axes, respectively, while maintaining sub-meter maximum position deviations throughout the evaluated trajectories. These results confirm that the proposed framework provides reliable localization and predictive state estimation for cooperative robotic navigation in indoor GPS-denied environments. Future work will investigate outdoor validation and extend the framework to additional data-driven decision-making models for future robotic services. Full article

Social media platforms have become major vectors for financial and cryptocurrency fraud, resulting in substantial economic losses and eroding user trust. This research presents a dual model system integrating machine learning-based user verification with deep learning-based content analysis to detect fraudulent activity more effectively than traditional single dimensional approaches. The core innovation lies in fusing these two modalities using a logical OR strategy. This design facilitates the detection of hybrid fraud schemes such as compromised legitimate accounts posting deceptive content or fake accounts spreading benign-looking messages that typically evade isolated detection systems. For user verification, ensemble methods were evaluated on a large, balanced dataset of social media profiles. Among the individual classifiers evaluated, the random forest classifier achieved the strongest performance and was selected for the final architecture due to its optimal balance of accuracy, interpretability, and computational efficiency. For content analysis, a convolutional neural network (CNN) trained on a substantial corpus of crypto-related posts demonstrated high accuracy, outperforming traditional keyword-based and recurrent neural network baselines. Ultimately, combining these models yields a system that flags significantly more fraudulent posts than either component alone. The user verification component retains interpretability through feature importance measures, while the CNN-based content analysis component operates as a less transparent classifier; the combined system can be adapted across diverse social media platforms. Full article

Parkinson’s disease (PD) diagnosis and severity assessment increasingly rely on gait analysis. To capture complex gait dynamics from limited sensor data, we propose a novel multi-branch framework integrating a one-dimensional convolutional neural network (1D-CNN), a bidirectional gated recurrent unit (Bi-GRU), and Kolmogorov–Arnold networks (KANs) (CNN-GRU-KAN) utilizing 18-channel vertical ground reaction force (VGRF) signals. Each module serves a distinct clinical purpose: the 1D-CNN branch with squeeze-and-excitation (SE) attention extracts localized spatial plantar pressure patterns, while the Bi-GRU branch with temporal attention captures long-range rhythm abnormalities. Crucially, the KAN serves as the classification head. By utilizing learnable B-spline functions instead of traditional fixed activations, KAN adaptively models the highly non-linear boundaries between healthy controls and varying PD severities, effectively mitigating overfitting. Under rigorous subject-independent cross-validation, our model achieves 98.43% accuracy for binary PD detection and 93.46% for five-class UPDRS severity grading. These results highlight the framework’s strong potential for low-cost, unobtrusive clinical tracking and home monitoring. Full article

Accurate estimation of the state of health (SOH) of lithium ion batteries is a fundamental prerequisite for the safe and reliable operation of battery management systems. To address the issues of insufficient feature representativeness, manually dependent hyperparameter tuning, and limited generalization under small sample conditions in existing SOH estimation methods, this paper proposes a hybrid SOH estimation approach based on a convolutional neural network and bidirectional gated recurrent unit optimized by the RIME optimization algorithm. Unlike existing CNN GRU/LSTM models that rely on unidirectional recurrent structures and can only utilize forward temporal information, the proposed CNN-BiGRU architecture captures bidirectional contextual dependencies inherent in battery degradation, enabling more comprehensive characterization of aging dynamics from limited cycle data. Firstly, 13 health indicators (HIs) related to capacity degradation are extracted from the incremental capacity (IC) curves, and Spearman’s rank correlation coefficient is employed to select the optimal feature subset with the highest correlation. Secondly, a CNN BiGRU hybrid architecture is constructed, where CNN extracts local degradation features and BiGRU captures bidirectional temporal dependencies. More importantly, instead of relying on manual trial and error or grid search for hyperparameter determination, the RIME algorithm is introduced to automatically and globally optimize the core hyperparameters of the model. Finally, ablation and comparative experiments are conducted on the public NASA dataset. The results demonstrate that the proposed model significantly outperforms other comparison methods in terms of MAE, MAPE, and RMSE for three battery cells under both 80% and 60% training set ratios, confirming its comprehensive superiority in estimation accuracy, robustness, and generalization capability with limited samples. Full article

Traffic sensors provide real-time measurements of current traffic conditions, whereas traffic management applications require forecasts of future traffic speed or flow. This study considers the incremental expansion of an operating sensor network, in which a pre-trained network-level forecasting model must predict traffic states at sensor locations that were absent during training. Many spatio-temporal graph neural networks rely on sensor-specific embeddings and graph connections learned from data-rich training networks. These representations are undefined for previously unseen locations, limiting the direct application of pre-trained adaptive-graph forecasters during sensor-network expansion. To address this cold-start problem, we propose support-conditioned sensor-adaptive meta-graph learning (SC-SAMG), which derives target-node representations and spatial dependencies from a short support period. The framework combines a support-set encoder, a task-specific graph learner, and first-order meta-learning to adapt the network-level forecaster using one to seven days of target observations. Experiments on four traffic benchmarks evaluate forecasts of speed or flow over the next 15–60 min under a leakage-controlled held-out-node protocol. SC-SAMG consistently outperforms fine-tuned gated recurrent unit (GRU), adaptive graph convolutional recurrent network (AGCRN), and diffusion convolutional recurrent neural network (DCRNN) baselines. It reduces mean absolute error (MAE) by up to 11% relative to the adaptive-graph baseline and by up to 7% relative to the diffusion convolutional baseline. These results demonstrate the potential of support-conditioned graph adaptation for incorporating previously unseen sensor locations into existing network-level traffic forecasting systems. Full article

Keyword spotting (KWS) is the always-on front end of voice interfaces; running it directly on microcontrollers, rather than streaming audio to the cloud, is essential for low-latency, privacy-preserving, and energy-efficient operation, yet it must reconcile high accuracy with severe memory, compute, and energy limits. Existing small-footprint KWS solutions are developed and benchmarked almost exclusively on English and are seldom validated on physical hardware or under quantization for other languages; their transferability to typologically different, consonant-rich languages such as Polish therefore remains largely unverified. We present a KWS pipeline for Polish, deployed and benchmarked on the RP2350 microcontroller (Raspberry Pi Pico 2, Raspberry Pi Ltd., Cambridge, Great Britain).We propose three compact architectures from the convolutional (CNN), convolutional recurrent (CRNN), and depthwise separable neural network (DS-CNN) families and benchmark them against the state-of-the-art BC-ResNet on a 25-keyword Polish vocabulary using MFCC features. All models are evaluated in full precision (Float32) and after eight-bit integer (INT8) post-training quantization, with inference latency measured directly on the target hardware. BC-ResNet attains the highest full-precision accuracy (97.81%) but is the most fragile under quantization, whereas the proposed CRNN is the most accurate quantized model (94.28%), the DS-CNN the smallest (46.15 KB), and the CNN the fastest (107.5 ms); all quantized models meet a one-second real-time budget. We further show that the accuracy ranking inverts after quantization, that memory savings are highly architecture-dependent, and that phonetically similar Polish words are the dominant source of error. These results offer practical guidance for deploying small-footprint KWS in Polish and other underrepresented languages. Full article

Surface soil moisture (SM) is the keystone variable of terrestrial ecohydrology. Yet, the rapid diversification and development of deep learning architectures for satellite SM estimation have outpaced practitioners’ capacity to select among them. This review synthesizes 37 deep learning studies from the SMAP era (2015–2026) across five architecture families (MLP and physics-informed neural networks [MLP/PINN], long short-term memory [LSTM] and gated recurrent unit [GRU] networks, convolutional neural networks [CNN], convolutional LSTM and graph neural networks [GNN], and Transformer-based models) to establish an architecture–task-matching framework that links each family to its dominant estimation niche. The analysis reveals consistent specializations: MLP/PINN models achieve competitive surface SM retrieval from satellite inputs; recurrent networks extend SMAP temporally (RMSE ≤ 0.035  ${\mathrm{m}}^3\hspace{0.17em}{\mathrm{m}}^{-3}$); CNN disaggregates SMAP to 1 km (reported unbiased root-mean-square error (ubRMSE) approaching 0.04  ${\mathrm{m}}^3\hspace{0.17em}{\mathrm{m}}^{-3}$); ConvLSTM and GNN address spatiotemporal gap-filling (low reported ubRMSE 0.022  ${\mathrm{m}}^3\hspace{0.17em}{\mathrm{m}}^{-3}$); and Transformers enable global multi-source fusion and decadal climate-scenario projection. Across all families, four physics-DL integration modes (hard architectural constraints, soft loss-function penalties, physics-as-input feature engineering, and physics-ML hybrid output fusion) consistently yield RMSE reductions of 8–50% relative to data-driven baselines. These findings provide a practitioner-oriented framework that is applicable to ecohydrological monitoring of plant water stress, agricultural drought, early flood warnings, and land–atmosphere coupling. Full article

Left atrium segmentation from magnetic resonance imaging (MRI) is essential for ablation planning in atrial fibrillation; however, clinical MRI quality is often degraded by noise, artifacts, and incomplete spatial coverage, making traditional recurrent neural networks (RNNs) vulnerable to such distortions. We developed a hybrid architecture inspired by cortico–cerebellar interactions to enhance segmentation stability without compromising mean accuracy. We utilized the open ATRIA dataset (100 patients, isotropic 3D MRI scans with manual left atrium annotations). The model comprises a convolutional encoder, a cortical RNN, and a cerebellar predictive module trained to predict future encoder features across multiple temporal horizons, generating a corrective feedback signal for the RNN. Experiments were conducted on unperturbed and degraded datasets with performance evaluated using the Dice coefficient. On unperturbed data, the cerebellar model achieved a mean best Dice of 0.835 ± 0.032 vs. 0.832 ± 0.027 for the baseline. Under degraded conditions, it showed significantly higher Dice (0.815 ± 0.019 vs. 0.801 ± 0.021; p = 0.014) and Surface Dice (p = 0.040), with a directionally lower between-run variance, though this difference in variance was not formally tested given the limited number of runs. nnU-Net achieved higher absolute accuracy but required three orders of magnitude more inference time and an order of magnitude more parameters. The cerebellar module improved boundary accuracy and reproducibility relative to the non-predictive baseline at a fraction of nnU-Net’s computational cost, offering a lightweight alternative for settings where deploying a full 3D self-configuring model is impractical. Full article

The increasing integration of renewable energy sources and smart grid (SG) technologies introduces significant challenges to power system stability due to the inherent variability and uncertainty of electricity generation and demand. While many machine learning and deep learning approaches have been proposed for stability prediction, various studies focus only on predictive performance and offer limited assessment of computational efficiency, statistical significance, and sustainability-related metrics. To address these gaps, this study suggests a hybrid fusion approach that combines Convolutional Neural Networks (CNNs), eXtreme Gradient Boosting (XGBoost), and Random Forest (RF) classifiers through a soft voting strategy for SG stability prediction. The CNN component automatically extracts representative features, while XGBoost and RF contribute complementary classification capabilities, reducing the need for manual feature engineering. In addition to predictive evaluation, a Green AI-oriented benchmarking framework is introduced to evaluate model performance using predictive accuracy, computational runtime, memory consumption, and estimated computational CO₂ emissions associated with model training and inference. The proposed framework is assessed on the UCI SG Stability dataset using stratified cross-validation and statistical significance testing, including Friedman and Nemenyi post hoc analyses. Experimental results demonstrate that the fusion model achieves 97.68% classification accuracy and an AUC of 0.991, exceeding several individual machine learning, deep learning, and ensemble baselines. Statistical analysis shows significant improvements over recurrent deep learning models such as LSTM and GRU. However, the differences from strong tree-based methods, including XGBoost and RF, are not statistically significant. Furthermore, the proposed model reaches a high sustainability score of 0.802, indicating a favorable balance between predictive performance and computational resource requirements. The findings show that the proposed framework is effective and computationally efficient to predict the stability of the smart grid on the UCI benchmark dataset and also serves as a transparent green AI benchmarking methodology for comparative studies in the future. The validation of the approach on real-world smart grid data under noisy, not fully complete, and heterogeneous operating conditions is another interesting research direction for future work. Full article

Living nervous systems sense the cardiovascular rhythm with a parsimony that engineered monitors cannot match: sensory receptors encode changes rather than absolute levels, neurons communicate through sparse all-or-none events, retinal circuits sharpen salient features through lateral inhibition, and attention is allocated by surprise. We translate these four principles into BioSpike-Net, a fully event-driven spiking neural network for cardiac-rhythm classification from electrocardiogram (ECG) and photoplethysmogram (PPG) signals. A sensory-adaptive spike encoder (SASE) converts analogue waveforms into ON/OFF spike trains through a mechanoreceptor-inspired gain-control law; adaptive-threshold leaky integrate-and-fire layers integrate these events; a lateral-inhibition spiking convolution emphasises locally salient morphology; and a novelty-gated temporal attention mechanism concentrates computation on the most surprising portions of each beat. Evaluated on the MIT-BIH Arrhythmia Database, PTB-XL, CPSC-2018, and a PhysioNet-derived PPG corpus, BioSpike-Net achieved 97.6 ± 0.3% accuracy and 95.8 ± 0.4% macro-F1 on MIT-BIH five-class arrhythmia classification, and 0.982 ROC-AUC on PPG atrial-fibrillation detection, matching or exceeding strong recurrent, convolutional, and transformer baselines while requiring an estimated 6.4 µJ per inference—approximately 27-fold below the transformer baseline—owing to a mean activation density below 0.10 spikes per neuron per time step. Ablations show that each biological principle contributes a measurable and interpretable accuracy-versus-energy benefit, and the network degrades gracefully under additive noise and motion artefact. By grounding architecture in the economy of biological sensing, this work offers a route to sustainable, always-on cardiac monitoring. Full article