Analysis of Bio-Signals for Drivers' Stress Level Detection

Abstract

In this study, the individual performances of the selected features, obtained from the ECG, GSR, EMG and RESP measurements by applying Pearson correlation analysis on the features accepted in the literature, were examined for the stress level detection. Accordingly, 2-, 1- and 3-dimensional feature sets were generated from ECG, Foot GSR and RESP measurements, respectively. These feature sets are classified by LLC, k-NN (k = 5), RF, DT and SVM algorithms. The feature set generated from the foot GSR measurement shows the best success with an accuracy of 66.67% when the LLC algorithm is used. This result indicates that the selected features are descriptive for stress level when they are used together.