IoT‑Enabled Multimodal Approach for Low‑Latency Prediction of Elite Athlete Performance Dynamics

The increasing deployment of wearable and neurophysiological sensing technologies in elite sports enables continuous monitoring of athletes’ cognitive, physiological, and biomechanical states; however, existing approaches often analyse unimodal data and fail to capture complex cross-modal interactions that govern performance dynamics. This research proposes a multimodal neurophysiological and kinematic data fusion framework for predictive modelling of elite athlete performance dynamics. Data collection was conducted using wearable EEG headsets, heart rate monitors for heart rate variability (HRV), and inertial measurement units (IMUs) to capture kinematic parameters during training and competition across multiple elite sports. The collected high-frequency time-series data were transmitted through an IoT infrastructure to an edge–cloud platform for real-time monitoring and analytics. Pre-processing included band-pass filtering used to signal denoising removal for EEG and physiological signals. The proposed method employs a hybrid deep learning architecture that integrates Temporal Variational Autoencoder with Vanilla Recurrent Neural Network (TVAE-Vanilla RNN) model to predict elite athlete performance. The Intelligent Biosensor Dataset used in this study was collected using wearable EEG headsets, heart rate monitors for HRV, and inertial measurement units (IMUs) to capture kinematic parameters during training and competition across multiple elite sports.