An effective software cross-project fault prediction model for quality improvement

As a quality assurance activity, cross-project fault prediction (CPFP) involves building a model for predicting the faults in a specific software project (aka target project) facing the shortage of within-project training data, leveraging cross-projects data. However, the quality of training data decides the success of a CPFP model. Existing CPFP approaches mainly focused on instance selection with minimal attention to feature selection. Further, the validation of their models has been performed through non-effort-based performance measures (NEPMs) only, considering the availability of unlimited inspection resources, which is impractical. Addressing these problems, we propose a Hybrid Training Data Selection (HTDS) approach combining feature selection along with instance selection for building a proficient and pragmatic CPFP model by validating its effectiveness in terms of effort-based performance measures (EPMs) along with the NEPMs to confirm its practical applicability. The empirical results on 62 datasets manifested the potency of the proposed approach over all the compared approaches in terms of NEPMs and EPMs collectively. Thus, our proposed HTDS approach results in a more productive and practical CPFP model that can empower practitioners to produce quality software at a lesser cost.(c) 2022 Elsevier B.V. All rights reserved.