Optimizing testing efforts based on change proneness through machine learning techniques

For any software organization, understanding the software quality is desirable in order to increase user experience of the software. When we talk about security software this factor becomes even more important. This paper aims to develop models for predicting the change proneness for object oriented system. The developed models may be used to predict the change prone classes at early phase of software development. Rigorous testing and allocation of some extra resources to those change prone classes may lead to better quality and it may also reduce our work at the maintenance phase. We apply one statistical and 10 machine learning techniques to predict the models. The results are analyzed from Receiver Operating Characteristics (ROC) analysis using Area under the Curve (AUC) obtained from ROC. Adaboost and Random forest method have shown the best result and hence, based on these results we can claim that quality models have a good relevance with Object Oriented systems.