Stress Detection And Alleviation Via Electrodermal Activity And Generative Music

Corradine, Christopher

Abstract

Accurate psychological stress detection systems have been created using a variety of methodologies and can provide users with real-time stress monitoring. Such systems can aid with providing early intervention and therapies for alleviation on order to chronic stress which is known to be detrimental to health. Previous research has shown music listening to be an effective form of stress alleviation and there is a wealth of knowledge regarding the associations between music parameters and induced emotional states. This work focuses on bridging the gap between these distinct areas to create a single system capable of detecting stress and using the resulting stress level to inform the generation of music for alleviation. A stress detection model has been created by training a random forest classifier on features extracted from samples of electrodermal activity measured during multiple affective states. MIDI data was then generated using a Markov model trained on a bespoke MIDI dataset, and musical parameters such as mode, velocity and tempo were modulated using the stress classification to apply the iso-principle. The resulting generative model is therefore a hybrid between stochastic and rule-based models. A proof-of-concept system has successfully been built along with footage of it functioning at the following link
https://www.youtube.com/watch?v=SMfPrT2OJ-o&ab_channel=ChrisCorradine. This work emphasises the need for higher resolution stress detection methods and makes suggestions for
a real-time system. The best performing stress detection model was subject dependent and achieved an accuracy of 86% and an F-Measure of 0.94.

• Thesis Type: Thesis
• Deposit Date: Oct 6, 2023
• Publicly Available Date: Oct 30, 2023
• Award Date: Sep 29, 2023