Wearable Biosensor Fusion with AI for Continuous Psychological Stress Monitoring

Abstract

This study’s primary goal is to quantitatively assess how wearable technologiessuch as smartwatches, biosensor wristbands, sports GPS trackers, and electromyographic sensorsaffect key physiological and psychological metrics, including heart rate, Heart Rate Variability (HRV), recovery rate, psychological stress levels, and stress management skills in professional Chinese tennis players. To address these issues a new model is developed Hybrid Multitask Deep Learning for Psychological Stress Detection (HMDL-PSD) model utilizes wearable sensor data to identify stress and emotions. Physiological signals were collected from fifteen subjects using the Empatica E4 wristband and RespiBAN chest belt. The Stress Detection Process involves data preparation, normalization, and deep learning training optimized for edge deployment using TensorFlow Lite.Sliding window techniques extract features from time-domain, frequency-domain, and nonlinear signals. The DL Classification module uses 1D Convolutional Neural Networks (1D-CNN), Long Short-Term Memory (LSTM), Transformer, Autoencoder, and Random Forest (RF) to identify temporal and complex patterns.Fusion Mechanisms use rulebased switching, Hidden Markov Model (HMM) fusion, and Online-HMM adaptation for physiological state description, with Bayesian filtering for reliable estimation under uncertainty.Advanced regularization improves generalization, while the Bayesian Uncertainty module evaluates confidence using variational inference and Monte Carlo Dropout (MCD). The Explainability Protocol enhances fairness through counterfactual analysis, meta-gating, and SHAP.Cortisol analysis validates efficacy in detecting stress-related hormonal changes. The experimental result is accuracy, precision, recall, F1-Score, HMDL-PSD calibration error, confusion matrix for stress, HMDL-PSD selective prediction.