Open Access

ARTICLE

A Feature Selection Method for Software Defect Prediction Based on Improved Beluga Whale Optimization Algorithm

Shaoming Qiu, Jingjie He, Yan Wang^*, Bicong E

School of Information Engineering, Dalian University, Dalian, 116622, China

* Corresponding Author: Yan Wang. Email:

Computers, Materials & Continua 2025, 83(3), 4879-4898. https://doi.org/10.32604/cmc.2025.061532

Received 26 November 2024; Accepted 21 February 2025; Issue published 19 May 2025

Abstract

Software defect prediction (SDP) aims to find a reliable method to predict defects in specific software projects and help software engineers allocate limited resources to release high-quality software products. Software defect prediction can be effectively performed using traditional features, but there are some redundant or irrelevant features in them (the presence or absence of this feature has little effect on the prediction results). These problems can be solved using feature selection. However, existing feature selection methods have shortcomings such as insignificant dimensionality reduction effect and low classification accuracy of the selected optimal feature subset. In order to reduce the impact of these shortcomings, this paper proposes a new feature selection method Cubic Traverse Ma Beluga whale optimization algorithm (CTMBWO) based on the improved Beluga whale optimization algorithm (BWO). The goal of this study is to determine how well the CTMBWO can extract the features that are most important for correctly predicting software defects, improve the accuracy of fault prediction, reduce the number of the selected feature and mitigate the risk of overfitting, thereby achieving more efficient resource utilization and better distribution of test workload. The CTMBWO comprises three main stages: preprocessing the dataset, selecting relevant features, and evaluating the classification performance of the model. The novel feature selection method can effectively improve the performance of SDP. This study performs experiments on two software defect datasets (PROMISE, NASA) and shows the method’s classification performance using four detailed evaluation metrics, Accuracy, F1-score, MCC, AUC and Recall. The results indicate that the approach presented in this paper achieves outstanding classification performance on both datasets and has significant improvement over the baseline models.

---

Keywords

---