PM-SRCANet: A Privacy-Preserving Multimodal Stress Recognition Convolutional Attention Network Model

Jichao Xiong, Wanxuan Wu, Jiageng Chen, Chunhua Su, Weizhong Zhao, Junyu Lin, Dian Jiao

Abstract: This study introduces PM-SRCANet, a privacy-preserving framework for stress recognition using multimodal physiological signals. With chronic stress linked to cardiovascular diseases and immune dysfunction, accurate detection systems are essential for health monitoring. PM-SRCANet addresses existing model limitations by combining convolutional neural networks with attention mechanisms to capture complex relationships in physiological data. The framework implements TFHE fully homomorphic encryption to ensure secure inference on encrypted data without compromising performance. Evaluated on the WESAD 3-class emotional classification dataset, PM-SRCANet achieved 99.72% accuracy, surpassing state-of-the-art methods while maintaining reasonable computational overhead for encrypted inference. This integration of privacy-preserving computation with advanced multimodal learning offers a secure, efficient solution for reliable stress recognition in privacy-sensitive applications.

External IDs:dblp:conf/wasa/XiongWCSZLJ25