Data Augmentation for Deep Neural Networks Model in EEG Classification Task: A Review.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Electroencephalogram (EEG) classification is crucial for brain-computer interface (BCI) systems, but extracting features from non-linear, non-stationary signals remains challenging.
• Deep Neural Networks (DNNs) offer automatic feature extraction but struggle with limited datasets, leading to overfitting and poor generalizability in practical BCI applications.

Purpose of the Study:

• To review and analyze various data augmentation (DA) strategies for improving DNN-based EEG classification.
• To investigate the impact of different DA methods on the performance and generalizability of EEG decoding models.
• To summarize current practices and outcomes to guide future research in DA for EEG classification.

Main Methods:

• A comprehensive review of recent studies on DA techniques applied to EEG classification using DNNs.
• Categorization of EEG-based BCI paradigms and the types of DA methods employed.
• Analysis of reported accuracy and performance metrics achieved with different DA strategies.

Main Results:

• Data augmentation (DA) is a promising technique to address the challenge of limited datasets in EEG classification.
• Various DA strategies have been developed and applied to enhance DNN models for EEG signal processing.
• The effectiveness of DA methods varies depending on the BCI paradigm and the specific DNN architecture used.

Conclusions:

• DA significantly improves the performance and generalizability of DNN-based EEG classifiers, especially when dealing with small datasets.
• The choice of DA strategy should be tailored to the specific BCI paradigm and the characteristics of the EEG data.
• This review provides insights into current DA practices and outcomes, offering guidance for future research and development in BCI technology.