Towards Physiologically-Responsive Interactive Garments with Machine Learning Techniques


Emotional experiences shape our lives every day. Negative emotions can impact not only our mood but also our biological signals, overall health, and wellness, especially if they are not addressed. Emotion-regulation and self-care techniques, such as meditation and exercise, can help to alleviate these emotions, but we have to remember to actively engage in them. Compression applied to the body, called Deep Pressure Stimulation, has also been shown to help suppress reactions from our nervous system under stress. In this work, we address the challenges of emotion regulation when experiencing negative emotions while doing desk work. To accomplish this, we custom-built two interactive jackets that have a removable, embedded microcontroller, sensors, and airbags. The airbags are used to apply compression to the sides when a user presses a button. 12 participants interacted with the jackets during a user study and were interviewed after. Data collected from 8 of these 12 participants during the user study was used to train 3 machine learning models: Logistic Regression, Support Vector Machine and XGBoost. Over 4 different testing conditions, XGBoost proved to be an efficient and effective predictor of when users choose to turn on the pump. Coupled with interview data from participants exploring desires for slow interfaces and automatic actuation, we have established a foundation for individualized interactive garment analytics using customized models for affect prediction.

MSc Thesis

Co-advised by