Stacking based approach for prediction of faulty modules

Determination of a software module, prone to fault is important before the defects are discovered; because it can be used for better prioritization of resources. Software fault prediction is one of such tasks that predicts the fault proneness of the developed modules by applying machine learning techniques on software defect data. State-of-art software defect prediction techniques suffer from achieving good accuracy due to the imbalanced nature of software defect datasets. To address this issue, here we present an approach for software defect prediction by combining imbalance removal and ensemble-model. As ensemble approach is very effective and provides better prediction results as compared to the individual techniques. The stacking-based framework is developed by considering the outperforming ensemble classifiers in order to predict the faulty software modules. All the experiments are performed over twelve benchmark NASA MDP datasets. The paper presents an improved ensemble-based stacking approach to classify the fault prediction for the software system in an effective way.