ECG and EEG features during stress

Mental health greatly affects the quality of life. The ability to detect and classify multiple levels of stress is therefore imperative. The aim of this work is to develop machine learning models for detection and multiple level classification of stress through ECG and EEG signals for both unspecified and specified genders. The models for the detection of stress from ECG are developed for real-world use, while the models based on ECG and EEG for the detection and multiple level classification of stress are devised towards clinical use. The detection models are achieved through developing and evaluating multiple individual classifiers. Stacking technique comprising multiple individual classifiers working in conjunction, on the other hand, is employed to achieve multilevel stress classification models. The models were trained and tested using ECG and EEG features extracted from 30 subjects. It is shown that most of the time kNN and SVM classifiers yielded highest classification accuracy for stress detection from ECG, and ECG together with EEG, and the performances of our models for unspecified (mixed) genders are comparable to those of other studies for both the detection and multilevel classification of stress. This study also reveals that the difference in genders affects classification accuracy for low-stress detection from ECG, any levels of stress detection from ECG and EEG, and multilevel stress classification from ECG and EEG – the classification accuracy for females is lower than that for males.