The sensor data fusion model combining 1D-CNN and BiLSTM achieved a classification accuracy of 89.47% and an F1-score of 0.90, outperforming single-modal EDA and PPG approaches.
A multimodal sensor data fusion model using 1D-CNN and BiLSTM improves the accuracy of student psychological stress classification compared to single-modal approaches.
Absolute Event Rate: 89.47% vs 82.15%
With the development of artificial intelligence and sensor fusion, student psychological stress classification is shifting from subjective assessment to intelligent recognition based on multi-source physiological signals.However, existing methods often rely on single-channel signals and insufficient temporal and multimodal modelling.This study proposes a stress classification method based on sensor data fusion and neural networks.Multi-source EDA, PPG, and body temperature signals are synchronously collected, denoised using biased Kalman filtering, and jointly modelled by 1D-CNN and BiLSTM to capture local and global temporal features.Experimental results show that CNN and BiLSTM achieve accuracies of 78.32% and 82.15%, respectively, while the fused model reaches 89.47% with an F1-score of 0.88.Multimodal fusion improves the F1-score to 0.90, exceeding single-modal EDA (0.70) and PPG (0.71).Kalman filtering reduces RMSE by 42.5% and increases entropy by 2.0 bits, enhancing classification performance and providing technical support for psychological stress management and intervention.
Qian et al. (Thu,) conducted a other in Psychological stress. Sensor data fusion and neural network (1D-CNN and BiLSTM) vs. Single-modal models (CNN, BiLSTM) was evaluated on Classification accuracy. The sensor data fusion model combining 1D-CNN and BiLSTM achieved a classification accuracy of 89.47% and an F1-score of 0.90, outperforming single-modal EDA and PPG approaches.