Solving the imbalanced class problem in software defect prediction using GANS

Abstract

Prediction of software defects is a highly researched and important domain for cost-saving advantage in software development. Different methods of classification using attributes of static code were used to predict defects in software.However, the defective instances count is very minimal compared to the count of non-defective instances and this leads to imbalanced data, where the ratio of data class is not equal. For such data, conventional machine learning techniques give poor results.While there are different strategies to address this issue, normal oversampling methods are different versions of the SMOTE algorithm, These approaches are based on local information,instead of the complete distribution of minority class.GANs is used to approximate the true data distribution of minority class data used for software defect prediction.