Yazılım Hata Tahmini için Sıralı Sınıflandırma Yaklaşımı

Abstract

Software bug prediction is the process of utilizing classification and/or regression algorithms to predict the presence of possible errors (or defects) in a source code. However, current classification studies in the literature assume that the target attribute values in the datasets are binary (i.e. buggy or non-buggy) or unordered, so they lose inherent order between the class values such as zero, less and more bug levels. To overcome this drawback, this study proposes a novel approach which suggests ordinal classification methods as a solution for software bug prediction problem. This article compares ordinal and nominal versions of various classification algorithms (random forest, support vector machine, Naive Bayes and k-nearest neighbor) in terms of classification performance on real-world 38 software engineering datasets. The results indicate that ordinal classification approach achieves better classification accuracy on average than the traditional (nominal) solutions.