All Times EDT
Classification of EEG Signals: An Interpretable Approach Using Functional Data Analysis (309895)Nedret Billor, Auburn University
*Yuyan Yi, Auburn University
Jingyi Zheng, Auburn University
Keywords: EEG, time-frequency analysis, functional test, B-spline basis, group LASSO,
Electroencephalography (EEG) is a typically noninvasive method to record electrical activity of the brain. In this paper, we aim to interpretably classify different human behaviors from EEG signals. Specifically, we propose a method to identify the regions of brain and predict human behaviors based on information from the EEG signals in time and frequency domain. Since EEG data is continuous flow of voltages, it can be considered as functional data, and functional data analysis(FDA), with advantage of interpretability, can be performed. Three-stage algorithm by using FDA is generated to obtain interpretable classification. Both time and frequency-related information will be extracted by transforming signals into time-frequency domain firstly. Next, feature selection is considered to reduce the dimension by functional testing. Finally, penalized multiple functional logistic regression model will be applied to interpretably classify specific human behaviors. Based on results of this method, we can interpretably classify the spatial distance from EEG signals with conclusion that the area of frontal and parietal with rhythm of delta and theta are more related to distance judgments.