Improving Machine Learning Models with Hybrid Metaheuristic Algorithm for Software Defect Prediction

Abstract

Software defect prediction has always been an area of interest in the field of software engineering. As the prediction of software defects plays a vital role, researchers are focusing more on metaheuristic algorithms to develop better prediction models. In this paper, we focused on the parameter tuning of the machine learning (ML) models using hybrid metaheuristic algorithms. Here, we have used three metaheuristic algorithms, namely sparrow search, wolf pack, and artificial bee colony optimization algorithm (ABC), to optimize the hyperparameters of the ML model. We have developed a hybrid version of these algorithms for better performance. The sparrow search algorithm (SSA) has high search accuracy and slow convergence speed with the advantages of good stability and strong robustness. The Wolf Pack Algorithm (WPA) has a robust global optimization ability, fast convergence speed, and various optimization strategies. The artificial bee colony (ABC) optimization algorithm has the advantage of not being influenced by the initial parameters, thus enabling search in a wider search space. Considering the strongest features of the aforementioned algorithms, two new hybrid algorithms have been developed, namely sparrow search algorithm-wolf pack algorithm (SSA-WPA) and sparrow search algorithm-artificial bee colony (SSA-ABC). These two algorithms are combined with the artificial neural network and XGBOOST model for better accuracy. To achieve the correctness of the proposed method, it is being verified by five defective NASA datasets and compared with the base methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.