Recognition predictive modeling using electroencephalogram

Abstract

A machine learning model that operates on raw electroencephalogram (EEG) signals is essential for accurately discerning the user’s current thoughts. Given the difficulty of categorizing EEG signals for use in brain-computer interface (BCI) programs, we adopted a systematic approach in this study to select an optimal predictive model. To enhance the effectiveness of our systematic approach, we extracted features such as band powers, averages, and root-mean-squared values. K-nearest neighbor (KNN), principal component analysis, and dual-layer neural networks were employed to evaluate and validate the effectiveness of the extracted features. The BCI IV competition-I dataset was utilized for analysis and validation. KNN achieved an average classification success rate of 98.02% compared to other methods. Furthermore, our research extends the application of this approach using it to create, test, and evaluate human driving behavior as a case study.