Revealing the Importance of Local and Global Interpretability in Smartphone Based Human Activity Recognition
V H Arul, P Karthikeyan, Elangovan Ramanujam · 2024
Human Activity Recognition (HAR) is the art of identifying and naming activities using Artificial Intelligence (AI) from the gathered raw activity data by utilizing various devices such as Smartphones, Vision systems, Wearable sensors, Ambient sensors, etc., The AI models that includes Machine Learning and Deep Learning techniques implemented towards the HAR is entirely black box model, and many users do not know the mathematics behind the process. Thus, the eXplainable Artificial Intelligence (XAI) model named Shapley Additive exPlanations (SHAP) is implemented in this paper to deal with the local and global interpretations of the Smartphone-based HAR dataset on how it achieves the predictions using an XGBoost classifier and a SHAP Tree explainer. The SHAP visualization plot makes the end-user to understand how each and individual feature/ complete feature set impacts the prediction of activities, either positively or negatively in real-time.