Selecting features of breathing & stability data to predict stress in students

Fischer, Johannes Simon

Selecting features of breathing & stability data to predict stress in students

License

Citation

Fischer, J. Selecting features of breathing & stability data to predict stress in students. 2021. handle: http://hdl.handle.net/10230/49221

This citation was generated automatically.

Abstract

Repeated or long-time exposure to psychological stress is a risk factor for mental and physical health. Especially students are exposed to stress and lacking stress management skills make them particularly vulnerable to stress-related illnesses. The Breathing Dynamic Modelling for Body Mind Interaction in Students (BYMBOS) project tries to tackle this problem by investigating the triangular relationship between breathing, posture, and psychological stress. This exploratory study contributes to this project by selecting discriminative physiological breathing and stability features that can detect stress. For this purpose, four feature selection methods were applied, namely a Genetic Algorithm, a Decision Tree, a Correlation-based, and a Relief-based feature selection approach. For the classification a Support Vector Machine (SVM), as well as the k-nearest Neighbor (KNN) classifier were used. Stability and respiratory data were recorded before and after relaxation was induced through deep breathing and muscle relaxation exercises. The application of the feature selection methods on the physiological data confirmed that including feature selection as a preprocessing step can not only reduce the number of variables significantly but can also increase accuracy, due to the elimination of noise. Predicting stress based on breathing and stability data achieved a leave-one-subject-out (LOSO) mean accuracy of up to 85.42%. Stability measures were found to boost the predictive power but were not sufficient for an effective prediction. However, these results may be biased as the relaxation inducing intervention included deep breathing exercises and several subjects were not able to perform deep breathing correctly. Hence, a different labelling procedure based on whether a person was able to perform deep breathing or not, was proposed. The prediction with this labelling solely based on stability measures achieved a LOSO mean accuracy of up to 87.76%. This suggests a relationship between stability, breathing, and psychological stress. Nonetheless, further research with more samples, and procedures to validate the perceived stress in subjects is required to support this hypothesis.