"Stress Classification Using Lightweight Approach with Single- and Continuous-Cycle ECG Fusion(Coming soon)"
Stress diagnosis typically involves using single- or continuous-cycle electrocardiogram (ECG) signals independently, which may fail to capture fine-grained beat level morphology or long-term temporal dynamics. In this study, we propose DeepBeat-Rhythm-FusionNet (DBRFNet), which integrates features from both single- and continuous cycle ECG signals. The proposed model achieves better performance than that achieved using either single or continuous-cycle ECG signals individually. ECG signals are classified using a framework that leverages parallel depthwise separable convolutions, an Adaptive Weight Sum Fusion mechanism, and a Transformer Encoder to effectively capture the morphological, relational, and temporal features corresponding to stress. The experimental results obtained using four public datasets—WESAD, MAUS, CLAS, and SWELL-KW—demonstrate that the proposed DBRFNet outperforms previous single- or continuous-cycle-based methods, achieving accuracies of 99.40%, 93.00%, 92.30%, and 93.07%, respectively, in multiclass classification tasks involving three or more stress-related categories. Additionally, real-time inference experiments conducted on a Raspberry Pi 4B demonstrated low latency (1.17–2.26ms on average) and high throughput (>300 cases/s). We demonstrate that the feature fusion incorporation of both single and continuous-cycle information enhances the diagnostic performance in ECG-based stress detection, demonstrating the potential of DBRFNet as a practical, scalable, and lightweight model for real-world deployment