Multi-Class Imbalanced Data Classification Using TwinSVM-One versus All and Synthetic Minority Over-sampling Technique

Imbalanced multi-class datasets present significant challenges in classification tasks within machine learning. This study introduces a novel approach by integrating Twin SVM-OvA with the Synthetic Minority Over-sampling Technique to improve classification performance on imbalanced datasets. The effectiveness of the proposed method was tested on six benchmark datasets, with performance evaluated using metrics including accuracy, precision, recall, and the F1-measure. The experimental results demonstrate that Twin SVM-OvA-SMOTE outperforms other baseline methods, achieving the highest average accuracy of 88%, with improvements ranging from 7.4% to 10.1%. The proposed method achieved perfect classification on specific datasets, such as Dermatology, and significantly enhanced performance on highly imbalanced datasets like Arrhythmia. These results highlight the effectiveness of the proposed method in addressing class imbalance and improving classification performance.