Feature selection using the zebra optimization algorithm for software fault prediction: a study on the bughunter dataset

Abstract

Software fault prediction focuses on identifying software modules that are most likely to contain faults before the testing stage, helping developers allocate quality assurance resources effectively and improve system reliability. A major challenge in SFP lies in redundant and irrelevant features within software fault datasets, which often lower the accuracy of predictive models. To address this, the study introduces a wrapper-based feature selection method using the Zebra Optimization Algorithm (ZOA). Experiments on nine BugHunter datasets show that the ZOA-based method consistently surpasses a baseline deep learning model trained on raw data, achieving higher F1-score, Precision, and Recall. The findings demonstrate that ZOA is effective in reducing feature redundancy and improving prediction performance. This research confirms the potential of ZOA in SFP, offering practical benefits for software development and opening new opportunities for further studies.