Predictive Modeling on MEG Signal to Classify Hand and Wrist Movement using UNEQ and KNN

The domain of Brain-Computer Interface (BCI) explores how humans can interact with the computer without giving direct instruction. Recognizing the activity from brain signals affiliated with an electronic device might (i.e. MEG) provide a stepping stone in the field of BCI. The intended algorithm in the paper aimed at presenting a statistical strategy to classify the brain signal from the MEG signal data, provided by BCI competition IV dataset III. The algorithm is compartmentalized in three levels: preprocessing, feature extraction, and classification. Autoregressive features have been extracted from the signals to classify using UNEQ, KNN and SIMCA, discuss the data distribution and asses how well the algorithm performs on unknown yet similar distribution. The proposed algorithm has obtained 64% prediction accuracy and 67% validation accuracy, which exceeds the current highest result reported on the same dataset.