Analysis of the Performance of Learners for Change Prediction Using Imbalanced Data

Software change prediction is important to economically schedule allocation of resources during various phases of software maintenance and testing. Furthermore, exact characterization of progress inclined and non-change inclined classes is significant in beginning times of programming advancement life cycle since that helps with creating financially savvy quality programming for real-time use. A good prediction model should predict both the change and non-change prone classes with high accuracy. However, most practical datasets have underrepresented information and serious class appropriation skews. Due to imbalanced data, the minority classes are not predicted accurately causing poor planning of resources. Popular operating systems like Android get updated very fast. In the current scenario, it is essential to recognize change prone and non-change prone classes with precision in newer versions of such software that are updated very frequently. In this paper, we give a complete survey of various machine learning models to predict change prone classes algorithms using sampling technologies like resampling and spreadsubsampling on six open source datasets having imbalanced data. The experimental result of the study advocates that resampling technique consistently and significantly improves the performance of all the models.