This project builds classification models that predict arm movement direction from human ECoG data. Patients with epilepsy moved arm to one of the four directions, while their ECoG signals were simultaneously (Fig. 6A). The project groups arm directions into two classes, left and right from the egocentric viewpoint of the patients (Fig. 6B). Diverse classification models are used and compared, among them linear SVM predicts the best with accuracy of 91.5%. Note that more complex models such as LSTM do not yield high performance, indicating that simply predicting arm movement direction might not need to capture complex temporal structures in ECoG.