Controlling a mobile platform using machine learning and EEG

Abstract

In recent years, the technology of Brain-Computer Interface (BCI) is gradually attracting the attention of researchers. The primary objective of BCI systems is to build a mutual connection between human brains and computers. It has great application potential in medical areas and might even change the way human lives. The analysis of BCI usually requires reliable electroencephalography signal classification techniques and feature extraction methods to obtain significant features from raw signals. This study aims to achieve an EEG classification system with the best performance by analyzing and comparing the different feature extraction and EEG classification methods. Multiple popular techniques including Short-Time Fourier Transform, Continuous Wavelet Transform, and Common Spatial Patterns, and algorithms such as Support Vector Machine, Multilayer Perception, and different neural networks are involved in the experiments based on the BCI Competition IV Dataset 2a. Conclusions are drawn that the algorithms of Common Spatial Pattern and Convolutional Neural Network perform well in this dataset. Furthermore, this study particularly focuses on the Filter Bank CSP algorithm and conducts further analyses to improve the performances. By the end of this study, an optimized system is proposed, which achieves an outstanding increase in the average classification accuracy to 0.7844.