Current Trends in Class Imbalance Learning for Software Defect Prediction
Somya Goyal
Abstract
Software defect prediction is of high importance to manage the software development efforts by focusing the testing efforts on the fault-prone modules. Imbalanced defect data causes detrimental impact on the performance of software defect predictors. Researchers deployed a diverse range of learning methods to tackle the class-imbalance issues resulting into the remarkable achievements improving the performance of defect prediction models. An analysis on the current trends addressing the Class Imbalanced Learning methods is essential in domain of SDP. This article presents a review of the contributions made over the period from year 2019 to 2024 towards handling class imbalance while applying learning machines for SDP. This study will provide current market trends to future researchers to handle class imbalance. This review has uncovered that – The most pronounced datasets are Promise and NASA which are highly imbalanced in nature. Hybridization of data-sampling techniques with ensembles is effective to handle the class imbalance issue. Among the available evaluation metrics, Area under the Curve is the most used one as it is insulated from the impact of imbalanced datasets. Deep learning models have potential and prospects to be explored for class imbalance handling in SDP in a full-fledged capacity.