Stress recognition based on the Markov Transition Field of Electrodermal Activity

Long-term psychological stress significantly impacts individual health, making stress recognition a critical area of research. To improve stress recognition accuracy, this study proposed a novel psychological stress recognition method using Electrodermal Activity (EDA). We first designed a data collection experiment using a sequential subtraction task, collecting EDA from 21 subjects. After preprocessing, EDA signals were converted into 2D images using the time-series to image encoding method, Markov Transition Field (MTF), and 10 texture features were extracted using Local Binary Patterns, a commonly used local textures describing method in image processing. Additionally, traditional multi-domain features were extracted. To optimize feature selection, three feature selection methods, Mutual Information (MI), Random Forest (RF), and Recursive Feature Elimination (RFE), were applied to the feature sets, and the selected features were input into a support vector machine for classification. The results showed that, for multi-domain features, the accuracies achieved using MI, RF, and RFE were 90.56%, 90.56%, and 97.78%, respectively. When MTF features were combined with multi-domain features, the recognition accuracies improved to 93.06%, 92.78%, and 98.50%, respectively. This study demonstrates that Markov analysis of EDA can provide additional valid information for stress recognition, further enhancing the utility of EDA in psychological stress assessment applications.