Translating Keystroke and Mouse Dynamics Data to Classify Human Mood

Many people struggle with mood disorders and breakdowns, such as depression, anxiety, and stress. These mood disorders spill over into the tasks that are completed throughout the day. For example, how a person interacts with a computer (e.g., how fast they type on a keyboard, move a mouse, and correct mistakes) can show the type of mood they are experiencing. In this study, we aim to contribute to the development of machine learning models to detect a person's mood in an unobtrusive manner. The detection of low/poor mood can contribute to the intervention of improving mental health. In particular, we collected more than 16,000 keystroke and mouse dynamics data. Next, we built two type-based (i.e., keystroke-based and mouse-based) models and evaluated six classification algorithms on our models. The results show that the mouse-based models performed the best results. The random forest (RF) and extreme gradient boosting (XGBoost) provided the highest overall scores for both keystroke and mouse dynamics. The classification models offer better accuracy for the pleasant mood than that of the unpleasant mood. Our findings indicate that mental states spill over into regular computer interactions. With the models we have developed, we contribute to the detection of human mood.