Browse Articles
Discover research articles across all indexed journals
LiteDriveNet for driver distraction classification using a lightweight multi-scale convolutional neural network
Comparison of Mosquito-Magnet trap and livestock-aspirated collection of mosquitoes and the implications for modelling mosquito-borne animal diseases
Automated tracking of the brown algal parasite Eurychasma dicksonii with deep learning
Harmonizing standards and resources for the medical genome
A TabNet-SHAP framework with stability-weighted multi-method feature selection for muscle injury prediction in professional football
Abstract Muscle injuries account for approximately 31% of all time-loss injuries in professional football, yet existing prediction models are constrained by small sample sizes, dependence on proprietary physiological data, limited interpretability, and inadequate temporal validation designs. This study develops an interpretable, strictly prospective prediction framework trained on 438,219 player-season records sourced from Transfermarkt, spanning 15 seasons (2010/11–2024/25) across 32 leagues. To ensure temporal integrity, all predictive features are constructed exclusively from prior-season data, and models are evaluated under a chronological hold-out design in which earlier seasons serve as training data and the two most recent seasons constitute the test set. We propose a multi-method stability-weighted feature selection (MSFS) strategy that cascades filter-based redundancy removal, four-method ensemble voting, and cross-validation stability verification, reducing 57 candidate features to a robust subset. Under cost-sensitive learning designed for severe class imbalance (6.2% positive rate), TabNet attains the highest discrimination among five candidate classifiers with an area under the receiver operating characteristic curve (AUC) of 0.812 (95% CI: 0.804–0.820), outperforming XGBoost (0.806), LightGBM (0.803), long short-term memory network (LSTM) (0.789), and logistic regression (0.761). Ablation experiments confirm that stability-verified feature selection and metaheuristic hyperparameter optimisation yield complementary performance gains, and direct comparison with Bayesian optimisation demonstrates that the detective behaviour algorithm provides competitive tuning across architectures. A seven-layer Shapley additive explanations (SHAP) interpretability analysis reveals that under the prospective design, injury history emerges as the most important feature category (33%), surpassing market-and-career features (29%), confirming that the temporal correction elevates genuine recurrence risk over reporting-related proxies. A two-stage risk architecture is identified in which market-and-career features establish a baseline risk stratum, whereas historical injury burden acts as a multiplicative amplifier among predisposed individuals. Individual-level decompositions further indicate that substitution-based rotation patterns are associated with attenuated age-related risk elevation. Sensitivity analyses excluding market value features quantify the extent to which predictive performance reflects genuine injury risk versus differential reporting completeness across leagues. These findings demonstrate that publicly accessible transfer-market data, combined with attention-based deep tabular learning and rigorous temporal validation, can support meaningful and interpretable injury risk stratification without proprietary monitoring systems.
Snoring classification with deep time-frequency features
Reliability of transversus abdominis and gluteus medius muscle thickness measurements using ultrasound imaging in chronic non-specific low back pain individuals
The complex truth about trust in science
Topology-oriented energy management systems based on supercapacitor charge sustaining and multi-port DC–DC converters for hybrid electric motorcycles
An EEG-based hybrid machine learning approach for CT scan triage in mild traumatic brain injury
Abstract Mild traumatic brain injury (mTBI) frequently prompts computed tomography (CT) imaging in emergency departments, despite a high proportion of negative findings. Objective, non-invasive tools that can support CT triage decisions under realistic clinical constraints are therefore needed. This study evaluates whether electroencephalography (EEG)-based biomarkers combined with temporal modeling can provide reliable decision support for mTBI assessment. Resting-state EEG was acquired using a clinically feasible 19-channel montage from 120 subjects classified as CT-Abnormal, CT-Normal, or healthy controls. Automated preprocessing was applied uniformly without manual artifact rejection. Quantitative EEG biomarkers were statistically validated and used to train a Random Forest classifier, while a Long Short-Term Memory (LSTM) network modeled temporal EEG dynamics. The biomarker-based model achieved a test accuracy of 81.25%. A hybrid fusion framework integrating Random Forest and LSTM outputs improved performance, achieving an accuracy of 93.33% and enhanced sensitivity for the CT-Normal category. Generalization was confirmed on an independent test set. These findings indicate that hybrid EEG representations can support CT triage decisions in mTBI as an adjunct to existing clinical assessment.
Scientists should recognize their own political biases to build public trust
Optimization of mechanical performance in bioinspired FDM-printed PLA sandwich structures using machine learning
Child online safety needs more than social-media bans
A nationwide cohort study of fracture incidence and postfracture mortality among older adults in Taiwan
Numerical investigation on the impact of solitary waves with variable incidence angles on offshore box-girder bridges
Correcting False Narratives — Indispensable Latino Contributions to U.S. Population Health
Sex differences in marathon pacing: analysis of 873,000 Berlin marathon runners reveals men are twice as likely to “hit the wall”
Toward Broader Adoption of Coronary Physiological Assessment
ML-based electromyography signal analysis for assessing rehabilitation exercise execution quality
Abstract Following lower limb trauma, performing orthopaedic rehabilitation exercises is a crucial factor in successful recovery. However, many patients find it difficult to execute these movements with optimal biomechanical form. To address this problem, an electromyography-based sensor system combined with machine learning is proposed. This system records and analyses exercises performed with the aim of providing real-time feedback on execution quality. To evaluate the potential of this concept, high-density electromyographic data (32 electrodes per sensor pad) were recorded from the vastus lateralis and vastus medialis muscles while performing rehabilitation exercises. Four different exercises (squat, hip abduction, leg raises and rocking) were performed, each in one optimal and three non-optimal variations, by n = 19 participants, resulting in 3,040 exercise executions and 194,560 electromyographic recordings. Analysis of this dataset demonstrated that a Support Vector Machine algorithm can be used to classify execution quality (four classes per exercise) with an average accuracy of 83.3% (± 8.8%). In addition, it was shown that, One Class Support Vector Machine trained with solely optimal executions, an unknown exercise could be identified as either optimal or non-optimal execution with an accuracy of 76.6% (± 5.9%). These results highlight the potential of this approach to evaluate exercise execution quality during rehabilitation. In the long term, this approach could provide personalised rehabilitation feedback and improve patient outcome.