P-dist Based Regularized Twin Support Vector Machine on Imbalanced Binary Dataset

Abstract

Data classification is a significant task in the field of machine learning. Support vector machine is one of the prominent algorithms in classification. Twin support vector machine is a solitary algorithm evolved from support vector machine which has gained popularity owing to its better generalization ability to a greater extent. Twin support vector machine attains quick training speed by explicitly exploring a pair of non-parallel hyperplanes for imbalanced data. In a Twin support vector machine, choosing numerical values for hyper parameters is challenging. Hyper parameter tuning is a prime factor that enhances the performance of a model. However, randomly preferred hyper parameters in the Twin support vector machine are uncertain. This paper proposes a novel p-dist-based regularized Twin support vector machine for imbalanced binary classification problems. Pairwise distances such as Jaccard and Correlation distances are considered for attuning the hyper parameters. The proposed work has been analyzed on many publicly available real-world benchmark datasets for both linear and non-linear cases. The performance of the p-dist-based regularized Twin support vector machine is computationally tested and compared with existing models. The outcome of the proposed model is validated using quality metrics such as Accuracy, F -mean, G-mean, and Elapsed time. Ultimately, the significant result exhibits better performance with less computational time in comparison to several existing methods.