Imbalanced biomedical data classification using self-adaptive multilayer ELM combined with dynamic GAN.

Area of Science:

• Biomedical data analysis
• Machine learning in healthcare
• Intelligent diagnosis systems

Background:

• Imbalanced data and limited samples are significant challenges in medical intelligent diagnosis.
• High-dimensional features in biomedical datasets negatively impact classification performance and disease diagnosis.
• Developing effective classification methods for such datasets is crucial but difficult.

Purpose of the Study:

• To propose a novel multilayer extreme learning machine (ELM) classification model integrated with dynamic generative adversarial networks (GAN).
• To address the challenges of limited samples and imbalanced data in biomedical datasets.
• To enhance the accuracy and robustness of medical intelligent diagnosis systems.

Main Methods:

• Utilized Principal Component Analysis (PCA) for feature extraction and reduction.
• Employed dynamic GAN to generate realistic minority class samples, balancing the dataset and preventing overfitting.
• Developed a self-adaptive multilayer ELM with analytically determined hyperparameters for robust classification.

Main Results:

• The proposed method successfully generated authentic minority class samples and adaptively selected optimal model parameters.
• Comparative experiments on four real-world biomedical datasets demonstrated superior classification performance.
• Achieved higher efficiency in ROC, AUC, G-mean, and F-measure metrics compared to existing methods (W-ELM, SMOTE-ELM, H-ELM).

Conclusions:

• The study offers an effective solution for classifying imbalanced and high-dimensional biomedical data with limited samples.
• The proposed method provides a theoretical foundation for computer-aided diagnosis.
• Demonstrated potential for application in clinical practice for improved medical diagnosis.