An Analysis of Different Variable Selection Methods in the Context of a Banking Institution

Variable selection has become important in some research areas such as data science and analytics, in which datasets with hundreds or even thousands of variables are what is available for model development. Among the objectives of variable selection, also called feature selection, is to improve the predictive capacity of input variables, as well as to provide faster and cost-effective predictors. A greater number of variables theoretically allow more information to be stored, but in practice this rarely happens due to the greater possibility of noise and redundancy in real-world data. Similarly, working with large data sets implies greater computational capacity and the principle of parsimony is no longer fulfilled. In this article, we use different methods for selecting variables, such as Gini, R², LASSO, and LARS to select variables in the context of predicting the probability of default of a mortgage loan database. A systematic way for variable selection is used taking advantage of the characteristics of the different selection methods.