Explainable Classification of EEG Data for an Active Touch Task Using Shapley Values

Machine learning has been used in the last decade to solve many problems in the haptics field. In particular, EEG data that is recorded during haptic interactions was used to train machine learning (ML) models to answer questions that are of interest to the neurohaptics community. However, the behavior of machine learning models in taking out their decisions is treated as black box hindering the interpretability of these decisions. In this paper, we used Shapley values, a concept from game theory, to explain the behavior of a tree-based classifier model in classifying electroencephalography data that was collected during an interaction with a surface haptic device under two conditions: with and without tactile feedback. We trained a tree-based ML model to classify data based on the presence or absence of tactile feedback. Using Shapley values, we identified the features (across and within channels) that contribute the most to the classification decision. Results showed channel AF3 and neural activity after 700 ms from the onset contributed the most in recognizing tactile feedback in the interaction. This study demonstrates the use of explainable machine learning in the field of Neurohaptics.