Classification of recorded EEG and EMG signals during various activities

In this thesis, it is aimed to classify the EEG and EMG dataset recorded simultaneously during 11 daily activities. This dataset, which includes measurements of 20 people, consists of one channel EEG data and one channel EMG data. Discrete wavelet transform was used to obtain the subcomponents of the EEG signals. In the next step, features of EEG, EMG, and sub-components of EEG signals were extracted. The features used in the study are mean absolute value, Hjorth mobility parameter, kurtosis, Hurst fractal index, and zero crossings. K-Nearest Neighbors (KNN), J48, Support Vector Machine (SVM), Artificial Neural Network (ANN), and Naive Bayes classifiers, which are widely used in the literature, were used in the classification step. EEG and EMG dataset and the dataset was obtained by combining these two data types were used for the classification. The highest accuracy with 94.7% was obtained using the KNN method from the dataset using EEG and EMG features. In addition, Principal Component Analysis was used to improve classification performance and reduce the noise ratio. Thanks to this method, the success obtained from the SVM method increased to 91.3%. Compared to other studies, the high accuracy classification of a multi-class dataset recorded with only two-channel electrodes and using the same features for EEG and EMG signals show the difference of this study.