Enhanced classification approach using MI-SVM for imbalanced multi-class datasets

Handling multi-class classification is inherently challenging, and the class imbalance problem exacerbates this complexity. Techniques tailored for binary imbalanced classification often fail to directly apply to multi-class scenarios, especially when both multiple majority and multiple minority classes are involved. This research introduces a novel approach, named MI-SVM, specifically designed for multi-class imbalanced datasets using support vector machines. MI-SVM addresses multi-class problems by effectively converting them into several binary sub-problems, constructing a hierarchical structure among all classes. A kernel transformation method is employed to tackle imbalance, which mitigates skew distribution within each binary class pair. This method reduces the computational complexity to a logarithmic scale and makes MI-SVM significantly faster than traditional multi-class SVM approaches. The validity of the proposed algorithm is demonstrated through experiments on various benchmark imbalanced multi-class datasets, showcasing its superior performance and efficiency.