A hybrid approach to software fault prediction using genetic programming and ensemble learning methods

Software fault prediction techniques use previous software metrics and also use the fault data to predict fault-prone modules for the next release of software. In this article we review the literature that uses machine-learning techniques to find the defect, fault, ambiguous code, inappropriate branching and prospected runtime errors to establish a level of quality in software. This paper also proposes a hybrid technique for software fault prediction which is based on genetic programming and ensemble learning techniques. There are multiple software fault prediction (machine-learning) techniques available to predict the occurrence of faults. Our experiments perform a comparative study of the performance achieved by simple ensemble methods, simple genetic programming based classification and the hybrid approach. We find that machine learning techniques have different learning abilities that can be exploited by software professionals and researchers for software fault prediction. We find that the performance obtained by this proposed approach is superior to the simple statistical and ensemble techniques used in the automated fault prediction system. However, more studies should be performed on lesser used machine learning techniques.