Selection of Learning Algorithm for Musical Tone Stimulated Wavelet De-Noised EEG Signal Classification

Abstract

The task of classifying EEG signals pose a challenge in the selection of which learning algorithm is best to provide higher classification accuracy. In this study, five wellknown learning algorithms used in data mining were utilized. The task is to classify musical tone stimulated wavelet de-noised EEG signals. Classification tasks include whether the EEG signal is tone stimulated or not, and whether the EEG signal is stimulated by either the C, F or G tone. Results show higher correct classification instances (CCI) percentages and accuracies in the first classification task using the J48 decision tree as the learning algorithm. For the second classification task, the k-nn learning algorithm outruns the other classifiers but gave low accuracy and low correct classification percentage. The possibility of increasing the performance was explored by increasing the k (number of neighbors). With the increment, its produced directly proportionate in accuracy and correct classification percentage within a certain value of k. A larger k value will reduce the accuracy and the correct classification percentages.