Özet:

The set of commands used to operate certain functions to operate computers and machines is called software. Today, activities and applications used in many fields contain software designed with different algorithms. It is of great importance that these softwares are perfect and in a way that meets the needs. The quality of the software and the absence of errors in the software are issues that both the developers of the software and the end users of the software attach importance to. Software defect prediction inherently involves an imbalanced class problem. In this study, first of all, the imbalanced class problem was tried to be solved. In this direction, different undersampling and oversampling methods were applied on the CM1, KC1, KC2, JM1 and PC1 datasets taken from NASA's PROMISE data repository, which is open to researchers in the literature. In the software defect prediction phase, different classification algorithms were compared and the most suitable algorithm was determined for each data set. In the experimental results, the imbalanced class problem in the datasets was resolved with ten different sampling methods; classification was done with thirteen different classification algorithms. With an AUC of 0.92, the best classification result was obtained in the PC1 dataset. With this study, it has been shown that the performance of defect prediction can be better with sampling methods and appropriate classifiers in software defect prediction. The results obtained were compared with the studies in the literature and the superiority and effectiveness of the proposed method were proven.

Anahtar Kelimeler: