Adaptive Bug Classification for CVE List using Bayesian Probabilistic Approach

Software bug classification is a precondition for bug fixation and it plays a vital role in software maintenance. It is found that bug fixation often takes long due to the distribution of misclassified or non-classified bugs by the triager among the developers. In this paper, we propose an adaptive bug classification approach on CVE dataset that involves two Bayesian classifiers such as Naive Bayes and Bayes net, and takes adaptive decision for classification. Naive Bayes is a classification algorithm which adopts a naive approach regarding class conditional distribution during classification and assumes that all the features of a sample are conditionally independent given the class label. Bayes net is a graphical representation of a set of random variables and their conditional dependencies via a directed acyclic graph. It can be used for knowledge representation as well as classification. Exploiting the domain knowledge about the bug class, we conduct the experiments from two different view points group-based approach and general approach. In case of groupbased approach, both classifiers are learned using the bug groupspecific samples and selected features from five groups. In case of general approach, 28,266 bug samples and 64 bug features are considered. Experiments show that Bayes net classifier has more potential than Naive Bayes for classification with CVE dataset and therefore it is preferable. However, it is also found that Naive Bayes classifier performs fairly well in CVE bug classification due to its simplistic concept and less number of parameters.