Electroencephalogram-Based Subject Matching Learning (ESML): A Deep Learning Framework on Electroencephalogram-Based...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram (EEG) signals reflect human brain activity.
• Accurate user identification and task classification from EEG are challenging.

Purpose of the Study:

• To introduce a novel deep learning framework, ESML (EEG-based Subject Matching Learning), for EEG-based user identification and task classification.
• To evaluate the effectiveness of ESML using raw EEG signals without preprocessing.

Main Methods:

• Developed ESML, a two-part deep learning framework: ESML1 (LSTM-based) for user identification and ESML2 (CNN-based) for task classification.
• Utilized raw EEG signals, eliminating the need for denoising or feature extraction.
• Validated the framework on three public EEG datasets.

Main Results:

• ESML demonstrated superior performance compared to traditional machine learning methods (SVM, LDA, NN, DTS, Bayesian, AdaBoost, MLP).
• ESML1 achieved 96% precision for user identification with 109 users.
• ESML2 achieved 99% precision for 3-class task classification.

Conclusions:

• The proposed ESML framework is effective and efficient for EEG-based user identification and task classification.
• Raw EEG signals can be directly utilized for these applications, simplifying data acquisition and processing.
• EEG signals hold significant potential for biometric identification and cognitive state monitoring.