Comparison of Meta-heuristic with Evolutionary and Local Search Methods for Feature Selection

Abstract

Feature selection is one of the most significant steps of pre-processing which deals with extracting the pertinent subset of features from the available feature set. This reduces the dimensionality of the dataset which is a key challenge in the field of data mining. In this chapter, three types of search methods are used for feature selection, viz. local search, evolutionary search and metaheuristic search. The goal of this chapter is to evaluate the search ability of metaheuristic methods as they overcome the drawbacks of local search and evolutionary search methods. For this, the authors have conducted an exhaustive comparison amongst the search methods where a collection of nine meta-heuristic search methods (Ant search, Bat search, Bee search, Cuckoo search, Elephant search, Firefly search, Flower search, Harmony search, Wolf search) are compared with three local search and four evolutionary search methods. These search methods are validated against two open source datasets belonging to different domains (allow to provide stronger and reliable conclusion), viz. Pima Indians Diabetes dataset and Hepatocellular carcinoma dataset found in UCI machine learning repository. After extracting the useful features using all the search methods, models are predicted using ensemble classifiers, Random Forest, Logitboost, Adaboost and Bagging. To the best of authors’ knowledge, there is no work in literature which has done such an exhaustive evaluation and comparison of search methods (total 16 search methods are used) and then constructed ensemble models to determine the effectiveness of each search method. The results proved that meta-heuristic search methods are more optimal for feature selection process and are better than local and evolutionary search methods.