Detection of Psychological Risk for Protected Individuals by Using PPG Signals from Smartwatch

Abstract

This paper proposes a machine learning approach to detect dangerous emition using short-term PPG (Photoplethysmogram) signals from a commercial smartwatch. In supervised learning, having accurately annotated training data is essential. However, a key challenge in emition detection problem is the uncertainty regarding how accurately data labeled as ”danger” reflects actual dangerous responses, since participants may react differently to the same experiments. The main contribution of this paper is the development of a feature selection method to remove ambiguously labeled training data, thereby improving the accuracy of the prediction model. In the test, PPG measurements were collected from participants playing a horror VR (Virtual Reality) game, and the proposed method validated the superiority of our proposed approach in comparison with other methods.