| Summary: | Feature selection or attribute reduction is a crucial process to achieve optimal data reduction for classification task. However, most of the feature selection methods that were introduced work individually that sometimes caused less optimal feature being selected, subsequently degrading the consistency of the classification accuracy rate. The aim of this paper is to exploit the capability of bio-inspired search algorithms, together with wrapper and filtered methods in generating optimal set of features. The important step is to idealize the combined feature selection models by finding the best combination of search method and feature selection algorithms. The next step is to define an optimized feature set for classification task. Performance metrics are analyzed based on classification accuracy and the number of selected features. Experiments were conducted on nine (9) benchmark datasets with various sizes, categorized as small, medium and large dataset. Experimental results revealed that the ideal combination is a feature selection model with the implementation of bio-inspired search algorithm that consistently obtains the optimal solution (i.e. less number of features with higher classification accuracy) on the selected dataset. Such a finding indicates that the exploitation of bio-inspired algorithms with ideal combination of wrapper/filtered method can contribute in finding the optimal features to be used in data mining model construction.
|
|---|