Motor Imagery for Brain-Computer Interface (BCI)

Real-time classification of motor imagery EEG signal using BCI competition II Dataset III to classify right & left-hand movement signals using matlab.

Preprocessing

• The Frequency response of the FIR bandpass filter (7 Hz - 15 Hz) using windowing method.

• Applying a Bandpass FIR filter on a window of the signal to extract the µ band

• Frequency Domain of the signal before and after filtration

Feature Extraction and selection

• For each window of the signal the energy of C3 and C4 channels is calculated using following equation

• Feature Space of the training data using different window length

Classification

• KNN classifier is used for classification with different values of K.

Evaluation Criteria

• Mutual Information (MI) and error rate (ERR)

Results

KNN	Max MI	Min Error %	Max Accuracy %
K = 1	0.64654	11.429	88.571
K = 3	0.82787	8.5714	91.429
K = 5	0.81647	10	90
K = 7	0.89775	8.5714	91.429
K = 9	0.82190	9.2857	90.714
K = 11	0.98919	7.8571	92.143
K = 13	0.78632	10.714	89.286
K = 15	0.89809	8.5714	91.429