An Unobtrusive Approach to Emotion Detection

Stress, a negative emotion state, is being pervasively experienced in mod- ern life, significantly impacting both physical and mental well-being, often going unnoticed until its effects become severe. Technological solutions for stress detec- tion offer a proactive approach to managing this pervasive problem. This paper in- vestigated a case study using web-based learning, an unobtrusive approach, to de- tecting stress. The solution involved the development of two Neural Network mod- els (ANN and CNN) to be deployed and processed on the client device through a Web application. For this, two features were developed (journal and selfie) and the models classified the user’s inputs as either ’stressed’ or ’not stressed’ in real time. Underpinning the work is Web learning, tensorflow.js, a framework devel- oped by Google. The application was evaluated by 7 users. Results: the application and models worked as intended, and an average score of 3.57 out of 4 for user’s subjective view on the accuracy was obtained. This result suggests the models were perceived as working well during user evaluation. The results also reviewed the users preferred the journal over the selfie feature. All users (100%) considered the features effective at monitoring and tracking stress.