Fig. 3: MLP classifier SHAP feature analysis.
From: 3D pose estimation for scalable remote gait kinematics assessment
The given figure shows a time-series representation of significant SHAP values represented in the form of a bounding box obtained from the MLP Classifier. The blue bounding box represents the negative impact of SHAP values concerning SCI classification probability and the red bounding box represents the positive impact. A, B represent hip angle/flexion and C, D represent knee angle/flexion time-series for first subject with SHAP values highlighting significant joint angle time-features essential for distinguishing between Healthy and SCI classes in Subset 2. These plots showcase cross-validated time-series features such as reduction in the range of joint camera frame angles, that were initially found in the K-Means Classifier quantified through SHAP values.
