Mouse Trajectories and State Anxiety: Feature Selection with Random Forest

Do users´ mouse activities reveal their affective states, as other bodily expressions such as postures and gestures signal emotions? When people are frustrated while trying to solve a puzzle or math problem, their frustration can be manifested in the way they use a computer mouse, such as pressing a button hard. But when a user is engaged in an innocuous and mundane task, what mouse activities provide a clue to detect affective states? To address these questions, we extracted 134 mouse trajectory variables in a choice-reaching experiment (N=234, female = 137, male = 97) and selected 3~8 key features by applying random forest regression. Using Spielberger´s State Anxiety Inventory, we investigated the extent to which the selected trajectory features predict state anxiety of new subjects (N = 133, female = 75, male = 58). Results indicate that distributions of temporal features (e.g., velocity) as well as spatial characteristics (e.g., direction change) are indicative of users´ state anxiety. A theoretical rationale, pros and cons of using mouse movement analysis and the role of other psychological variable for mouse-based affective computing are also discussed.