Software defect prediction using global and local models

There are certain crucial areas that require attention in the field of software defect prediction, even after extensive research. One of these issues that need attention is data heterogeneity. To address the issue of heterogeneity, local models have garnered attention. Limited studies have proven local models to be better than global models, so there is contradiction among researchers. Several researches carried out about using feature selection as a way to lessen the impact of heterogeneity but were not up-to mark. To resolve the issue of heterogeneity, feature selection technique and local models were used. Presented work shows a hybrid feature selection strategy (HFSS) with global and local (GL) models of software defect prediction.The work compared proposed approach with baselines techniques from literature namely DPDF and ANN on three PROMISE projects and traditional global models. Proposed technique achieved accuracy 10% higher for JM1, 6% more for PC3 dataset in comparison to baseline technique DPDF. In terms of f-measure, proposed technique performed 4%, 5% and 3% better for KC1, JM1 and PC3 dataset respectively than another baseline study ANN. Empirical results showcase that local models have preferential results than global models. The proposed HFSS has additionally improved the predicting power of GL models. Our proposed approach achieved better results in terms of accuracy, precision, recall and F-measure.