Improved support vector machine classification for imbalanced medical datasets by novel hybrid sampling combining...

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Imbalanced datasets pose challenges for machine learning models, particularly Support Vector Machines (SVMs).
• Existing sampling techniques may generate synthetic data ineffectively for SVMs, limiting classification accuracy.
• Improving SVM performance on imbalanced data requires novel approaches to handle class disparities.

Purpose of the Study:

• To propose a novel hybrid sampling technique, modified mega-trend-diffusion-extreme learning machine (MMTD-ELM), for imbalanced datasets.
• To enhance the classification accuracy of SVM models for minority classes.
• To effectively adjust the SVM decision boundary towards the majority class region.

Main Methods:

• Developed a hybrid sampling technique combining the α-cut fuzzy number method and the MMTD method.
• Screened representative majority class examples and created new minority class examples.
• Utilized a bagging ELM model to monitor the similarity between synthetic and original data.

Main Results:

• The MMTD-ELM method demonstrated superior performance across four datasets.
• Comparative analysis showed MMTD-ELM outperformed three state-of-the-art sampling techniques.
• Statistically significant improvements were observed in geometric mean (G-mean), F-measure (F1), index of balanced accuracy (IBA), and area under the curve (AUC).

Conclusions:

• The proposed MMTD-ELM method effectively addresses imbalanced datasets for SVM classification.
• MMTD-ELM offers a significant advancement over existing sampling techniques for improving minority class prediction.
• This novel approach provides a robust solution for enhancing SVM performance in scenarios with class imbalance.