An Ensemble Oversampling Model for Class Imbalance Problem in Software Defect Prediction

Software systems are now ubiquitous and are used every day for automation purposes in personal and enterprise applications; they are also essential to many safety-critical and mission-critical systems, e.g., air traffic control systems, autonomous cars, and SCADA systems. With the availability of massive storage capabilities, high speed Internet, and the advent of Internet of Things devices, modern software systems are growing in both size and complexity. Maintaining a high quality of such complex systems while manually keeping the error rate at a minimum is a challenge. Therefore, automated detection of faulty components in a software system is important during software development and also post-delivery. Fault detection models usually needs to be trained on a labeled-balanced dataset with both faulty and non-faulty samples. Earlier work, e.g. Mohsin et al. (2016), showed that most real fault detection training dataset are imbalanced. Thereby, the trained model gets over-fitted and classifies faulty components as non-faulty components. The consequence of a high false negative rate is cumulative and results in generating more errors when using the model in other software systems –never seen before, which is very expensive. In this paper, we propose a software defect prediction ensemble model which considers the class imbalance problem in real software datasets. We use different oversampling techniques to build an ensemble classifier that can reduce the effect of low minority samples in the defective data. The proposed approach is verified using PROMISE software engineering datasets. The results show that our ensemble oversampling technique can more greatly reduce the false negative rate compared to the standard classification techniques and identify the faulty components more accurately resulting in a less expensive detection system (lowering the rate of non-faulty predictions of faulty modules).