HSoMLSDP: A Hybrid Swarm-Optimized Machine Learning Software Defect Prediction Framework

Abstract

Defect prediction plays a crucial role for any software system across various domains, as its failure may cause unavoidable and undeniable scenarios. For reliable software, defect-free is considered as one of the most important criteria. This research aims to design a hybrid swarm-optimized machine learning software defect prediction (HSoMLSDP) framework to predict software defects. We strive to do this by designing a swarm-optimized machine learning defect prediction (SoMLDP) model within the HSoMLSDP framework. In pursuit of enhancing the defect prediction accuracy of the SoMLDP model, this paper introduces a hybrid swarm optimization algorithm (SOA) referred to as the gravitational force Lévy flight grasshopper optimization algorithm-artificial bee colony (GFLFGOA-ABC) algorithm. By combining the enhanced exploration feature of the gravitational force Lévy flight grasshopper optimization algorithm (GFLFGOA) with the robust exploitation capacity of the artificial bee colony (ABC), the GFLFGOA-ABC algorithm is proposed. Prior to validating the HSoMLSDP framework, the LFGFGOA-ABC algorithm's performance is first confirmed by experiments on 6 benchmark functions (BFs) to assess its mean and convergence rate. Following BF verification, the second experiment tunes the hyperparameters of ML models (ANN, GB, XGB) to improve the defect accuracy of the SoMLDP model. As an enhancement of accuracy justifies the correctness of the SoMLDP model, thus validating the HSoMLSDP framework. © 2025 IEEE.