Cross-Domain Classification Model With Knowledge Utilization Maximization for Recognition of Epileptic EEG Signals

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Epileptic EEG signal classification requires large labeled datasets.
• Conventional models struggle with data from different distributions (e.g., varying patient groups or acquisition devices).
• This limitation hinders the generalizability and robustness of existing classification methods.

Purpose of the Study:

• To develop a novel cross-domain classification model, CDC-KUM, for enhanced epileptic EEG signal recognition.
• To leverage both labeled source domain data and unlabeled target domain data.
• To improve classification performance when training and testing datasets have different data distributions.

Main Methods:

• The CDC-KUM model maps data into kernel space.
• It incorporates a pairwise constraint regularization term using labeled source domain data.
• It utilizes a soft clustering regularization term with quadratic weights and Gini-Simpson diversity for unlabeled target domain data.

Main Results:

• The CDC-KUM model demonstrated superior performance compared to traditional non-transfer and transfer learning methods.
• Experimental results validated the model's effectiveness in recognizing epileptic EEG signals across different data distributions.
• The model successfully utilized global data structure and distribution information from both domains.

Conclusions:

• The proposed CDC-KUM model offers a robust solution for epileptic EEG signal classification, particularly in cross-domain scenarios.
• Knowledge utilization maximization is key to overcoming data distribution discrepancies.
• This approach enhances the reliability of EEG-based epilepsy diagnosis systems.