GGO-GA: A Hybrid Group Geese Optimization–Genetic Algorithm for Efficient Feature Selection in Software Defect Prediction

Software defect prediction (SDP) plays a crucial role in modern software engineering by identifying potentially faulty modules prior to release, thereby reducing maintenance costs and improving product quality. However, high-dimensional and noisy feature spaces often hamper the performance and efficiency of prediction models. In this work, we introduce GGO-GA, a novel hybrid feature selection algorithm that marries the exploration–exploitation dynamics of the Group Geese Optimization (GGO) with the genetic operators of a classic Genetic Algorithm (GA). GGO-GA dynamically partitions the population into exploration and exploitation subgroups, adaptively adjusting their sizes to escape local optima, while GA’s crossover and mutation enrich the search for diverse feature subsets. We conduct extensive experiments on 18 real-world defect prediction datasets drawn from AEEEM, ReLink, NASA, and PROMISE repositories. Compared against four state-of-the-art methods (PCA, GWO, GAFS, EMWS), GGO-GA achieves on average a 21.0% improvement in G-measure and 19.8% uplift in AUC across all datasets. Wilcoxon and Cliff’s \(\delta \) analyses confirm the statistical significance and practical impact of these gains. The results demonstrate that GGO-GA can substantially reduce feature dimensionality while maintaining or improving prediction accuracy, offering an effective and efficient solution for SDP tasks.