Machine learning models combining Hybrid feature selection with kNN best distinguished non- and low-stress, while Filter feature selection with kNN best distinguished non- and high-stress conditions.
Machine learning models using EEG and ECG signals, specifically combining feature selection with kNN algorithms, can effectively detect stress levels for potential use in real-world applications.
The quality of life is greatly affected by mental health, and the ability to detect stress is imperative. The aim of this work is to develop machine learning models for stress detection through EEG and/or ECG signals with the capability to be used in real-world applications, namely smartphones and edge devices. This is achieved through developing and evaluating 12 machine learning models which combine 3 feature selection methods and 4 classification algorithms to detect stress. The models were trained and tested using EEG and ECG features extracted from 20 subjects. It is shown that the best, most practical machine learning models for distinguish non- and low-stress conditions is the combination of the Hybrid feature selection method and the kNN classification algorithm, and for distinguish non- and high-stress conditions is the combination of the Filter feature selection method and the kNN classification algorithm.
Hemakom et al. (Wed,) conducted a other in Stress (n=20). Machine learning models (feature selection and classification algorithms) was evaluated on Stress detection (distinguishing non- and low-stress, and non- and high-stress conditions). Machine learning models combining Hybrid feature selection with kNN best distinguished non- and low-stress, while Filter feature selection with kNN best distinguished non- and high-stress conditions.