Abstract: Feature selection (FS) plays a crucial role in pre-processing machine learning datasets, as it eliminates redundant features to improve classifcation accuracy and reduce computational costs. This paper presents an enhanced approach to FS for software fault prediction, specifcally by enhancing the binary dwarf mongoose optimization (BDMO) algorithm with a crossover mechanism and a modifed positioning updating formula. The proposed approach, termed iBDMOcr, aims to fortify exploration capability, promote population diversity, and lastly improve the wrapper-based FS process for software fault prediction tasks. iBDMOcr gained superb performance compared to other well-esteemed optimization methods across 17 benchmark datasets. It ranked frst in 11 out of 17 datasets in terms of average classifcation accuracy. Moreover, iBDMOcr outperformed other methods in terms of average ftness values and number of selected features across all datasets. The fndings demonstrate the efectiveness of iBDMOcr in addressing FS problems in software fault prediction, leading to more accurate and efcient models.

Key words: Dwarf mongoose optimization algorithm , · Optimization , · Feature selection , · Classifcation