Time-Series Classification of Multichannel EEG Signal Data using Statistical Methods and 1D-CNN

Classification of neuroimaging signals is an important but challenging task in biomedical research where the diagnosis and treatment of brain diseases is usually of interest. The aim of this study is to obtain an in-depth understanding of neuroimaging data by applying various machine learning and deep learning methods for classification of encephalopathy. We focus on the high-dimensional EEG signal where the study subjects are labeled by alcoholism and control. After some pre-processing steps of the matrix data using the Quantile-aggregation method, we conducted a comparison between the performance of machine learning methods and the 1D-CNN deep learning models where the entire matrix data including the time-series features as the input. As a result, 1D-CNN model showed a more stable performance with an 8.4% improvement in accuracy (0.956% +/-0.007) than the best ML model (Logistic Regression, 0.872% +/- 0.037). This study provides an open source dataset and is expected to serve as a useful resource for further research by visualizing the performance of several models for EEG data.

datasets from : https://archive.ics.uci.edu/ml/datasets/eeg+database

Sungjae Lee (sungjae.me) / Juyeon Lee ([email protected])
Prof. Weining Shen