Human Stress State Analysis from Foot Movements with Deep Learning

Abstract

In this study, 9-axis sensor data were used to classify people's stress levels. Given the significant effects of stress on psychological and physiological health, the development of such a classification system is of critical importance. The data collection process was carried out by having the participants play the Tetris game. During the game, as the user's level increases, the game speed and the probability of making errors also increase; this leads to an increase in the stress levels of individuals. By analyzing how foot movements change during stressful moments in individuals, a more in-depth understanding of stress is provided. The developed classifier models were created with Support Vector Machine (SVM), Long Short-Term Memory (LSTM) and AutoEncoder (AE) algorithms. Sensor data were examined for classification after a detailed preprocessing and feature extraction process. Performance analysis showed that the models gave quite successful results in stress classification. The results provide an important basis for the development of data augmentation and fusion methods for classes with different sample numbers. © 2025 Elsevier B.V., All rights reserved.