Stress Detection using CNN Fusion

Early stress detection has a significant impact on recognizing health issues. The aim of this paper is to discuss a Convolutional Neural Network (CNN) based multimodal fusion using a Multimodal Transfer Module (MMTM) to identify stress in a subject-independent way. Electrodermal Activity (EDA) and Electrocardiogram (ECG) physiological modalities are used for fusion. The performance of the proposed model is evaluated in comparison with and without transfer learning on the two benchmark datasets - CLAS and ASCERTAIN. Results show that multimodal fusion with transfer learning performance is higher than without transfer learning experiments. Furthermore, fusing modalities at a higher level of the network helped to enhance the model's efficiency.