Increasing Classification Accuracy of Motor Imagery EEG Signals with Logical Combination of Classifiers and by Applying Genetic Algorithm and Small Decision Trees

In this paper we present a two-step method to improve classification accuracy of EEG signal. The main objective of this paper is to improve the classification of motor imagery derived from brain signals. In this regard a hybrid classifier based on Boolean rules and genetic algorithm is presented that uses the features of time-frequency domains for feature extraction of EEG signal which contains statistical and non-statistical indicators obtained from the wavelet packet transform. In this paper in order to improve the classification results, in the first step a set of classifiers with different errors is created. At this point the extracted features are given to the decision tree classifier as base classifier. In the second step using genetic algorithms, optimal combination rule to combine the results of the classifiers is obtained. Combination rule is proposed according to the Boolean rules. For required data, third data set from second version of BCI competition data sets is used. Implementation results of the proposed method have shown accuracy of 96.43% which compared to the existing methods in EEG signal classification, have 6.43% better performance.