Temporal Analysis of Stress Classification Using QRS Complex of ECG Signals

This paper demonstrates that electrocardiogram (ECG) signals can be used to detect and classify stress in a person using as low as 5 s data stream. The analysis focuses on determining the minimum ECG data required to classify stress levels. Time taken to detect level of stress can be crucial to prevent cardiac arrest and other abnormalities. The ECG data of 10 drivers driving through different traffic conditions is segmented into 60, 30, 20, 15, 10, and 5 s instances. Two levels of stress, low and high, and features from Q-, R-, and S-fiducial points and classifiers such as Naive Bayes, logistic, multilayer perceptron, SMO (sequential minimal optimization), IB1 (nearest neighbor), J48 (decision tree), and random forest are used for experiments. The results show that stress can be identified with high accuracy. It is found that even a 5 s data stream provides an 87.98% accuracy using random forest twofold cross validation test, opening the door for rapid stress detection.