Software Fault Prediction using Bio-Inspired Algorithms to Select the Features to be employed: An Empirical Study

In recent past, the use of bio-inspired algorithms got a significant attention in software fault predictions, where they can be used to select the most relevant features for a dataset aiming to increase the prediction accuracy of estimation techniques. The most-earlier and widely investigated algorithms are Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). More recently, researchers have analyzed other algorithms inspired from nature. In this paper, we consider GA and PSO as baseline/benchmark algorithms and evaluate their performances against seven recently-employed bio-inspired algorithms and metaheuristics, namely Ant Colony Optimization, Bat Search, Bee Search, Cuckoo Search, Harmony Search, Multi-Objective Evolutionary Algorithm, and Tabu Search, for feature selection in software fault prediction. We present experiments with seven open source datasets and three estimation techniques: Random Forest, Support Vector Regression, and Linear Regression. We found that it is not always true that the recently introduced algorithms outperform the earlier introduced algorithms