Electroencephalogram-Based Subject Matching Learning (ESML): A Deep Learning Framework on Electroencephalogram-Based Biometrics and Task Identification

Authors: Jin Xu, Erqiang Zhou, Zhen Qin, Ting Bi and Zhiguang Qin

Year: 2023

Journal / Conference: Behavioral Sciences

Paper Link: https://doi.org/10.3390/bs13090765

Abstract:
An EEG signal (Electroencephalogram) is a bioelectric phenomenon reflecting human brain activities. In this paper, we propose a novel deep learning framework ESML (EEG-based Subject Matching Learning) using raw EEG signals to learn latent representations for EEG-based user identification and task classification. ESML consists of two parts: the ESML₁ model, which uses an LSTM-based method for EEG-user linking, and the ESML₂ model, which uses a CNN-based method for EEG-task linking. The ESML model is simple, effective, and efficient. It does not require any restrictions for EEG data collection related to user motions or thoughts, and it avoids EEG preprocessing operations such as denoising and feature extraction. Experiments on three public datasets demonstrate that ESML outperforms baseline methods (SVM, LDA, NN, DTS, Bayesian, AdaBoost, and MLP), with the ESML₁ model achieving a precision of 96% across 109 users and the ESML₂ model achieving 99% precision for 3-class task classification. These results provide strong evidence that raw EEG signals can be effectively used for user identification and task classification.