An Analysis of EEG Signal Classification for Digit Dataset

Abstract Artificial Intelligence (AI) and Machine Learning has brought significant atten- tion to the human brain, making it a prominent research area in engineering and technology and other non-medical sciences. Electroencephalogram (EEG) are one of many biological signals that are produced by human brain. EEG signals contain electrical properties and has frequency ranging between 0-100Hz. Fea- tures are the various attributes of the recorded signals which are associated with the state of the human brain. The data comprises values that correspond to the frequencies of EEG signals, specifically delta, theta, alpha, beta, and gamma. Additionally, it includes information about the level of attention, level of medita- tion, and the frequency of eye blinking of the subject. This research has given a notion of how a imagined digit is classified from an EEG signal by using machine learning algorithms. We have done the analysis by using models like k-Nearest Neighbor (kNN), Convolutional Neural Network (CNN) and Genetic Program- ming (GP). An original EEG data set is created for digit from 0–9 by using a non invasive single electrode (channel) EEG device. The obtained accuracy for kNN is 66.8%, for Convolutional Neural Network it is 73.1% and that for GP it is calculated equal to 82%. If the calculated accuracy of lower channel device is improved and further achieved more then one day they may replace higher chan- nel bulky devices. As single channel or lower channel EEG device are portable and easy to use therefore implementation of this work in future may meet a variety of applications in biomedical engineering, smart health care, personal assistance and automation.