Motivation
The core motivation behind this research stems from the growing recognition that emotions significantly influence human biomechanics, particularly gait and balance. Traditional emotion detection methods—like facial expressions or voice analysis—are easily manipulated or obstructed, whereas gait offers a subtler and more authentic window into emotional states. Understanding these biomechanical shifts not only provides a new pathway for emotion recognition but also holds critical clinical potential. For instance, gait-based emotion detection could serve as an early marker for mental health conditions, such as depression or bipolar disorder, where emotional fluctuations often manifest subtly in movement before becoming clinically apparent. Moreover, these insights could be applied to fall prevention strategies in vulnerable populations, as emotions like fear or sadness directly affect dynamic balance.
Our Solution
To tackle this challenge, the team utilized 3D motion capture systems and advanced machine learning models to classify emotions based on gait biomechanics. Over 150 biomechanical variables were extracted, representing joint angles, spatiotemporal parameters, and center-of-mass kinematics. Using algorithms like Random Forest, Logistic Regression, Multi-layer Perceptron (MLP), and XGBoost, the project explored which features most strongly correlated with specific emotions. Through feature selection and class balancing (SMOTE), classification accuracy improved, with XGBoost achieving nearly 60% accuracy—well above random chance. This validated the hypothesis that gait, when analyzed with machine learning, can predict emotional states. The study’s design—such as leave-one-participant-out cross-validation—ensured that results generalized beyond individual-specific movement patterns.
My Contributions
I was involved in multiple phases of the project, including experimental setup, data collection, and analysis. I helped place reflective markers for motion capture, administered emotion induction protocols, and assisted in managing trial sessions to ensure consistent data quality. On the computational side, I worked on preprocessing gait data, calculating biomechanical parameters, and organizing the dataset for machine learning input. I also contributed to running model training pipelines, tuning hyperparameters, and analyzing performance metrics.
