Heart Rate Variability-Based Mental Stress Detection Using Deep Learning Approach

Abstract

Health problems are rising with today’s stressful life, as it promotes cardiac diseases, depression, violence, and may provoke suicide. Hence, it is essential to develop a computer-aided diagnosis system to identify relaxed versus stressed individuals and their correct classification. Heart rate variability (HRV) based on RR interval is a well-proven clinical and diagnostic tool strongly associated with the autonomic nervous system (ANS). In this study, a conventional method was compared with a deep learning-based method. In the Conventional method, features were extracted from various domains, and these features were fed to a classifier to detect stressed states. However, this method uses hand-crafted features, and hence, there is a possibility of missed high potential features that may be responsible for maximizing the classifier’s generalization performance. This work presents a new approach motivated by the long short-term memory network (LSTM) in sequence learning to generate a concrete decision about the signal category. We proposed deep learning-based Inception-LSTM network to improve performance and to reduce computational cost. Two different stress datasets, viz., self-generated stress data and Physionet driver stress data were used to perform the proposed method’s performance analysis. The presented Inception-LSTM architecture outperforms existing literature methods, achieving an accuracy of 93% for self-generated stress data and 97.19% for driver stress data.