A Software Defect Prediction Classifier based on Three Minimum Support Threshold Association Rule Mining

With the increasing complexity of software system, the cost of software maintenance is increasing. In this case, software reliability is difficult to guarantee. To address this problem, software defect prediction technology based on machine learning has been attached great importance by a large number of scholars. Because of the strong interpretability of association rules, association rule algorithms are often used in classification tasks. However, the class imbalance problem seriously impacts the performance of traditional software defect classifiers based on association rule mining, therefore, it is necessary to use association rule algorithm that can be used to handle class imbalance data to deal with this problem. In this paper, a software defect prediction classifier based on three minimum support threshold association rule mining is proposed, which aims to improve the quality of these three frequent item-sets by considering the support of frequent item-sets containing defect labels, including non-defect labels and only including software metrics. The algorithm is compared with other four machine learning algorithms, and the results show that the algorithm is effective.