A Support Vector Machine Based Approach for Effective Fault Localization

Abstract

Software maintenance is one of the most costly activities in software life cycle. It costs almost 70% of the total cost of the software. Testing aims to reveal the faults from the software under test (SUT). The fault localization is tiresome, dull, costly but crucial for program debugging. As size and complexity of software increase, manual locating faults becomes very tedious and hence necessitates automatic fault localization. If the fault proneness of the software components can be predicted, then such components may be given more focus. Such approach would not only save time but also enhance the quality of the software. Support vector machine (SVM) is a prominent machine learning algorithm. Regularization of parameters, convex optimization and kernel tricks are the prevailing features of SVM. This work reports a SVM-based framework for fault localization on the basis of code smells. Paper presents a performance analysis against four popular algorithms, namely ZeroR, OneR, Naive Bayes and Decision Stump. The proposed model is empirically evaluated in the reference of Json project. The results of the experimentation show that the proposed model can effectively classify the instances in the classes of their respective categories of code smells. Also, the kernel used in the proposed model gives better performance than counterpart kernels and the proposed model itself performs better than the other compared algorithms in terms of accuracy, precision, recall and F-measure.