Maximum Margin of Twin Spheres Support Vector Machine for Imbalanced Data Classification.

Area of Science:

• Machine Learning
• Computational Intelligence
• Data Science

Background:

• Twin Support Vector Machine (TSVM) offers faster learning than standard Support Vector Machine (SVM) by solving two smaller Quadratic Programming Problems (QPPs).
• TSVM has limitations including an expensive matrix inverse operation and reduced effectiveness with imbalanced datasets.

Purpose of the Study:

• To propose a novel Maximum Margin of Twin Spheres Support Vector Machine (MMTSSVM) for improved imbalanced data classification.
• To enhance computational speed and overcome the matrix inverse limitation inherent in TSVM.

Main Methods:

• MMTSSVM formulates the classification problem by finding two homocentric spheres, one capturing majority class samples and the other enclosing minority class samples.
• The method involves solving a single QPP and a linear programming problem, avoiding the pair of QPPs in TSVM and the matrix inverse operation.

Main Results:

• MMTSSVM demonstrates significantly increased computational speed compared to SVM and TSVM.
• Experimental results on nine benchmark datasets show MMTSSVM's effectiveness against state-of-the-art algorithms for imbalanced data.
• The proposed MMTSSVM achieved superior results when applied to an Alzheimer's disease medical experiment.

Conclusions:

• MMTSSVM is an efficient and effective algorithm for imbalanced data classification.
• The novel approach overcomes key limitations of TSVM, offering a faster and more robust alternative.
• MMTSSVM shows promise for real-world applications, including medical data analysis.