Classification of EEG Data using k-Nearest Neighbor approach for Concealed Information Test

In this paper, an EEG based Concealed Information Test is developed. EEG is an acquisition technique of brain signal from brain scalp using electrodes. The main task here is to classify the EEG data into innocent and guilty. Data acquisition of 10 subjects has been carried out. The signal preprocessing is performed by passing the raw EEG signals through a band-pass filter. From these preprocessed EEG signals, it is necessary to extract significant features. In the time domain, the extraction of the statistical parameters such as mobility, activity and complexity is done from the EEG signals. The binary classification of the guilty and innocent classes in performed using k-nearest neighbor classifier. In order to validate the deceit identification system, 5-fold cross validation has been applied on the each of the subjects. To validate the performance of the classifier, performance measures such as accuracy, sensitivity, and specificity are taken into consideration. Out of three Hjorth parameters, mobility yielded better classification accuracy of up to 96.7%.