Feature Selection Algorithm for Data with Both Nominal and Continuous Features

Wenyin Tang, Kezhi Mao

Abstract: Wrapper and filter are two commonly used feature selection schemes. Because of its computational efficiency, the filter method is often the first choice when dealing with large dataset. However, most of filter methods reported in the literature are developed for continuous feature selection. In this paper, we proposed a filter method for mixed data with both continuous and nominal features. The new algorithm includes a novel criterion for mixed feature evaluation, and a novel search algorithm for mixed feature subset generation. The proposed method is tested using a few benchmark real-world problems.