An l1-Norm Regularized Copula Based Feature Selection

Abstract

In this paper, we develop a novel feature selection method called RCFS (Regularized Copula based Feature Selection) based on regularized copula. We use l1 regularization, as it penalizes the redundant co-efficient of features and makes them zero, resulting in non-redundant effective features set. Scale-invariant property of copula ensures good performance in noisy data, thereby improving the stability of the method. Three different forms of copula viz., Gaussian copula, Empirical copula, and Archimedean copula are used with l1 regularization. Results prove a significant improvement in the accuracy of the prediction model than any non regularized feature selection method. The number of optimal features to achieve a fixed accuracy value is also less than any other non regularized feature selection techniques.