Non-Linear Principal Component Analysis And Classification Of EEG During Mental Tasks

This thesis explores the effectiveness of Non-Linear Principal Component Analysis (NLPCA) as a technique for reducing the dimensionality of human electroencephelogram (EEG) for enabling it to be classified into two different metal tasks. EEG signals from a single subject recorded through six channels was studied during the performance of two mental tasks. An NLPCA network was used to reduce the dimensionality of temporal windows of eye-blink free EEG data. A standard backpropagation network was used to classify the reduced dimensionality representation of the original data. The results indicate that the NLPCA method is able to extract distinguishing features from the data that could be classified as belonging to one of the two tasks with an average percentage accuracy of 86:22%. The NLPCA method is found to be a definite improvement over its linear counterpart, the Karhunen-Lo`eve transform,..