Automatic Detection of Anger Based on Gait and Balance From Wearable Mobility Sensors

Anger affects muscle tension, cognitive performance, and overall well-being. Gait and balance are possible indicators of many emotional states. Computer models can be used to study their relationship with anger. This project intends to construct Machine Learning (ML) algorithms that can identify anger from gait and balance data, improving our understanding of anger's physical manifestations. One hundred seven participants completed the Profile of Mood Survey-Short Form (POMS-SF) to assess anger. Afterward, participants performed a modified Clinical Test for Sensory Interaction on Balance (mCTSIB) and a two-minute walk on a <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{6 m}$</tex> track wearing APDM mobility sensors. The collected data were utilized to optimize and train various ML models using decreased feature sets. To validate the models, an ANCOVA was performed to account for age, height, and weight. A support vector classifier using the top five features effectively classified people by anger with 85.3 % accuracy, demonstrating the potential of gait and balance data to determine emotional states. This study shows that gait and balance metrics can identify anger, opening new paths for physical and mental health interventions. Clinical relevance- By identifying anger, gait balance metrics can effectively indicate emotional states and pave the way for novel physical and mental health interventions.