Feature selection for software defect prediction using an improved firefly algorithm

Abstract

As all we know, the performance of the data-based learning models heavily depend on the quality of data that feature selection (FS) method is used to reduce the redundant features and irrelevant features to improve the classification performance. In this research, a feature selection method using an improved firefly algorithm (FA) is proposed to address the "curse-of-dimensionality" problem in software defect prediction (SDP). In the proposed method, where simulated annealing (SA) approach is used to avoid falling into local traps, a multi-group strategy is employed to obtain a promising optimal solution to the optimization problems, called the multi-group simulated annealing firefly algorithm (MSAFA). To evaluate the performance of MSAFA, excessive experiments have been conducted on a total of 28 optimization benchmark problems (CEC2013) and 6 class-imbalanced datasets selected from the NASA repository. In the comparison with other state-of-the-art swarm-based intelligence algorithms, MSAFA outperforms them to provide the optimal solutions to most optimization problems and SDP models. In short, the proposed feature selection method can obtain the promising solution to support data-based classified models in SDP