Analysis of Sampling Approach to Balance Data for Change Prediction

Software keeps on evolving due to the changing requirements and demands of the customer. Due to this, nowadays, we have multiple versions of a software. Incorporating any change at a later phase in the software development life cycle leads to investment of lots of resources. In this study, we will develop models to predict the classes which are more prone to changes at a very early stage in software development life cycle. This will allow the developers to focus their attention and resources on limited number of classes rather than all the classes. In addition, rigorous verification activities can be focused on such classes leading to less number of bugs/errors at a later stage. However, the datasets in software engineering domain suffer from the class imbalance problem leading to inaccurate or biased model prediction. In this study, we have used a sampling technique to balance the dataset. The aim of this study is to use meta heuristic algorithms for model prediction using imbalanced and balanced data (obtained after sampling). The empirical validation is conducted using a popular operating system, Android. The study compares the efficiency of different algorithms in predicting the change proneness and further shows that the performance of the models is comparable or better on the balanced data as compared to the imbalanced data.