EFG Signal Processing and Analysis Using Efficient Machine Learning Techniques
The purpose of this research is to explore electroencephalogram (EEG) signal features and evaluate the performance of different prediction models. This paper presents the methodologies of how to set up experiments to collect and visualize EEG data, discover feature correlations and build efficient classification models. Several experiments are conducted to collect raw EEG data from ten healthy subjects. In the experiment, subjects are asked to watch a looping video which shows rock, paper, and scissors repeatedly 90 seconds for each, they are also imagining doing the same thing in their mind. At the same time, a device called Emotiv Epoc placed on their head will record the EEG signals and transmit them to the computer. After data preprocessing and feature extraction, EEG data is fed into several different prediction models including Support Vector Machines, Logistic Regression, K-nearest Neighbors, and Neural Networks. The performance of the classification process due to different methods is presented and compared based on their accuracy, recall, precision, and F-1 score. The best model achieved in this study is Neural Networks with an accuracy of 76.5%, this is more than twice higher than guessing randomly with an accuracy of 33.3%. A GUI is also built as a brain-computer interface powered by the model with the best performance, which can be further developed for medical use or other purposes.