Combining CNN with Autoencoder for Motor-Imagery Classification

Motor Imagery (MI) is a significant component to start with Brain-Computer-Interface (BCI) research. For the past decade, EEG (Electroencephalogram) signal-based MI classification remains a major area of interest for researchers. In this paper, the accuracy of the multi-class classification of MI with EEG signal using machine learning (One-vs-One and One-vs-Rest), and Deep Learning (CNN - Convolutional Neural Network) using pre-processing techniques such as Wavelet Transform (WT) and Short-Time Fourier Transform (STFT) has been compared. The EEG signals are acquired from the PhysioNet Database (EEG Motor Movement/Imagery Dataset Version 1.0). From the comparative analysis study, the Deep Learning approach for Wavelet Transformed EEG signal gave the highest overall accuracy of 96.68% and AUC score of 0.87.