Fault prediction by utilizing self-organizing Map and Threshold

Abstract

Predicting parts of the programs that are more defects prone could ease up the software testing process, which leads to testing cost and testing time reduction. Fault prediction models use software metrics and defect data of earlier or similar versions of the project in order to improve software quality and exploit available resources. However, some issues such as cost, experience, and time, limit the availability of faulty data for modules or classes. In such cases, researchers focus on unsupervised techniques such as clustering and they use experts or thresholds for labeling modules as faulty or not faulty. In this paper, we propose a prediction model by utilizing self-organizing map (SOM) with threshold to build a better prediction model that could help testers in labeling process and does not need experts to label the modules any more. Data sets obtained from three Turkish white-goods controller software are used in our empirical investigation. The results based on the proposed technique is shown to aid the testers in making better estimation in most of the cases in terms of overall error rate, false positive rate (FPR), and false negative rate (FNR).