Evaluation of Spatio-temporal Deep Learning for Hand Movement based on Motor Imagery Electroencephalography

Electroencephalography (EEG) in the Brain-Computer Interface (BCI) system is used to process signal data and translate it into a user context in this research case Motor Imagery (MI). MI data requires further learning, especially on features for EEG data classification. The evaluation method carried out and the comparison in this test is the use of spatio-temporal methods to analyze EEG signal patterns on each channel used and signal patterns that change over time. The EEG data used is SHU-Dataset which has data from 32 channels and data collection for 72 minutes on 25 subjects with each session is 5 sessions and 2 classes. The spatio-temporal methods used in SHU_Dataset testing, namely; SVM with CSP and FBCSP extraction features, FBCNet, DeepConvNet, EEGNet, EEGNex, and ATCNet. From the results on testing spatio-temporal methods, the best results are in the EEGNex and ATCNet methods with average accuracy results on EEGNex is 75.67% and on ATCNet 76.05%.