Informative Software Defect Data Generation and Prediction: INF-SMOTE

Highly imbalanced data typically make accurate predictions difficult. Unfortunately, software defect datasets tend to have fewer defective modules than non-defective modules. Synthetic oversampling approaches, namely SMOTE, address this concern by creating new minority defective modules to balance the class distribution before a model is trained. Despite its success, these approaches come with the following shortcomings such as 1) over-generalization problem and generate near-duplicated data instances (less diverse data) due to oversampling of noisy samples, and 2) increasing the overlaps between different classes around the class boundaries. This paper introduces INF-SMOTE (Informative- Synthetic Minority Oversampling Technique), a novel and efficient synthetic oversampling approach for software defect datasets, simultaneously targeting all the shortcomings. INF-SMOTE identifies the informative minority samples that are appropriate for over-sampling. The process is in two way 1.) it identify and remove the noisy and overlapping samples from borderline minority instances based on the sampling seeds, and 2) synthetic samples are generated from the informative minority samples. Experiments were conducted on 12 releases of SDP (Software Defect Prediction) Datasets from the NASA repository. By comparing with the state-of-the-art techniques, we observe that the INF-SMOTE improves the defect prediction performance.