Wrapper-Filter Feature Selection using Multi-Objective Forest Optimization Algorithm: A Fusion Methodology

Abstract

The prevailing data upsurge has empowered the use of data for enhanced decision making process. Prior processing of data is a vital task to certify that the appropriate traits are measured during the study. To reduce the data extents, Feature Selection (FS) is an essential stage in the data pre-processing process along with stabilizing the significant data features. This paper introduces a Hybrid Wrapper Filter Multi-Objective Forest Optimization Algorithm with Local Search model (HWF-MOFOA-LS) algorithm for FS problem. The FS is performed using the hybrid wrapper-filter approach which are optimized concurrently. Initially a Multi-objective approach simultaneously optimizes the hybrid wrapper-filter fitness functions. The aim is to reduce the number of features and identify the related information to improve the classification accuracy. Next, the classified population with Pareto front solutions are enhanced by applying a local search selection strategy. Finally, the performance of the Multi-objective proposed technique are performed on 12 datasets from UCI repository. The results of the proposed approach is optimal when compared to the other multi-objective techniques. The proposed algorithms outperforms the other techniques by reducing the classification error along with selecting ¬minimum features.