Trial-based Classification of Haptic Tasks Based on EEG Data

With the increasing popularity of neural imaging techniques such as electroencephalography (EEG), developing quantitative measures to characterize haptic interactions is becoming a reality. Meanwhile, machine learning is a promising approach for trial-based EEG data analysis. This work presents a model that can distinguish between passive and active kinesthetic interactions based on a single trial EEG data. An interactive task that involves hitting a ball using a racket is developed under passive and active kinesthetic settings using a haptic device and a computer screen. Temporal and frequency domain features are extracted from the motor and somatosensory cortices, and a proposed 2-D CNN model is trained on data extracted from 19 participants. The model achieves a mean accuracy of 84.56%, 93.96%, and 95.89% across 5-fold validation when using one, four, or six electrodes, respectively. The model mechanism is assessed using an explainable machine learning algorithm, LIME, which shows that the model utilizes sensible features from a neuroscience perspective towards its prediction. This work paves the way for a better understanding of the neural mechanisms associated with kinesthetic haptic interaction, which proves helpful in many applications such as motor rehabilitation and brain-computer interactions, in addition to modeling the haptic quality of experience objectively.