Software defect prediction using global and local models

Abstract

Despite intense investigation in the area of software defect prediction, there are some critical regions that still need attention. Heterogeneity of data is one of these areas that seek attention.  Local models have gained focus in resolving the problem of heterogeneity. Limited studies have proven local models to be better than global models, so there is contradiction among researcher. Various researchers also considered feature selection as a method to mitigate the affect of heterogeneity. Our study presents a hybrid feature selection strategy with global and local (GL) models of software defect prediction (SDP). The proposed Hybrid Feature Selection Strategy (HFSS) has additionally improved the predicting power of GL models. Empirical results showcase that local models have preferential results than global models. Our study compared proposed approach with baselines techniques from literature on three PROMISE projects and traditional global models. Our proposed approach achieved better results in terms of accuracy, precision, recall and f-measure.