Fuzzy least squares projection twin support vector machines for class imbalance learning

In this paper, we propose a novel fuzzy least squares projection twin support vector machines for class imbalance learning (FLSPTSVM-CIL). Unlike twin support vector machine (TSVM) which solves two dual problems, we solve two modified primal formulations by solving two systems of linear equations. The proposed FLSPTSVM-CIL model seeks two projection directions such that the samples of two classes are well separated in the projected space. To avoid the singularity issues, we incorporate an extra regularization term to make the optimization problem positive definite. As the real world data may be imbalanced, we assign appropriate fuzzy weights to the samples such that the classifier is not biased towards the samples of the majority class. The statistical analysis and experimental results on the publicly available UCI benchmark datasets show that the proposed FLSPTSVM-CIL performs better as compared to the baseline models. To show the applications of the proposed FLSPTSVM-CIL model on real world datasets, we performed classification of Alzheimer's disease and breast cancer patients. Experimental results show that the generalization performance of the proposed FLSPTSVM-CIL model for the classification of the breast cancer patients and the mild cognitive impairment versus Alzheimer's disease subjects is better as compared to the baseline models.