FEATURE SELECTION AND PARAMETER OPTIMIZATION FOR SUPPORT VECTOR MACHINES USING PARTICLE SWARM OPTIMIZATION AND HARMONY SEARCH

The present paper proposes a mechanism, Diverse Particle Swarm Optimization and Harmony Search (DPSO_HS), which finds feature subsets and parameter values for Support Vector Machines (SVM) when addressing classification problems by incorporating Particle Swarm Optimization (PSO) and Harmony Search (HS). Specifically, we introduced HS to enhance diversity in the PSO process since it has the advantage of providing diverse solutions as compared to other methodologies, as it considers all solutions in memory when improvising a new solution. For performance evaluation, various datasets with a wide range of features, instances, and classes were considered. DPSO_HS showed an increased diversity and classification accuracy as compared to PSO where statistical significance was found in most datasets. In addition, with two different hybridized approaches based on PSO, we observed that the proposed method showed higher accuracy for most datasets. We also reviewed the results of previous research with identical datasets and found that DPSO_HS achieved higher or equal accuracy rates for most datasets.