Difference of Convex Functions Optimization for Feature Selection in Granular Ball Support Vector Machine

Resumo

Feature selection constitutes a critical optimization problem within the domain of supervised pattern classification. It involves selecting an optimal subset of features that maximizes the retention of the data’s salient information. Granular Ball Support Vector Machine (GBSVM) has proven to be a powerful technique for enhancing the predictive accuracy and computational tractability of classification models, by exploiting the concept of granular structures in the feature space, through the generation of a set of granular balls, enabling complex decision boundary modeling and adaptability to data variability. This paper presents a novel embedded feature selection approach in the context of granular ball SVM, directly enhancing classifier performance. Our approach to the resulting optimization problem is to apply Difference of Convex (DC) functions programming to effectively handle the non-convex nature of the problem. Genetic algorithm is used to tune the model’s parameters. Experimental results on UCI datasets show the efficiency of the proposed method.