Feature Selection in Mental Stress Analysis using Multiple Biological Signals

Maria Goñi, Inma Mohino, Cosme Llerena, Roberto Gil-Pita, and Manuel Rosa


Stress, feature extraction, genetic algorithm, linear discriminant, boostrap techniques


Stress is a response of people to face up to daily mental, emotional or physical challenges. Continuous monitoring of stress levels of a subject is of key importance to understand and control personal stress. In this sense, different biological signals can be used, such as, heart rate (HR), respiration, galvanic skin response (GSR) or electric response of the muscles. In this paper we extract a large number of features from the aforementioned biological signals in order to classify the levels of stress. Once we calculate these features, we use a genetic algorithm combined with a least square linear discriminant (LSLD) in the aim of selecting the most suitable features, considering the error of classification. Results show that respiration is the most useful signal in the classification of stress level and specifically, entropy and recurrence analysis of that signal are the most relevant features. In the case of GSR, we observe that feet are more sensitive to changes of the electrodermal activity than hands. With respect to EMG, it is the less adequate signal to classify stress level.

Important Links:

Go Back