A Survey and Comparative Study of Filter and Wrapper Feature Selection Techniques

Abstract

Feature selection is considered as a problem of global combinatorial optimization in machine learning, which reduces the number of features, removes irrelevant, noisy and redundant data. However, identification of useful features from hundreds or even thousands of related features is not an easy task. Selecting relevant genes from microarray data becomes even more challenging owing to the high dimensionality of features, multiclass categories involved and the usually small sample size. In order to improve the prediction accuracy and to avoid incomprehensibility due to the number of features different feature selection techniques can be implemented. This survey classifies and analyzes different approaches, aiming to not only provide a comprehensive presentation but also discuss challenges and various performance parameters. The techniques are generally classified into three; filter, wrapper and hybrid