An Evaluation Approach for Selecting Suitable Defect Prediction Method at Early Phases

Abstract

One of the most important quality indicators of a software product is its defect rates and accordingly, Software Defect Prediction (SDP) has become a widely-accepted practice in both academia and industry to reduce the cost of defects. Despite being reported less frequently, applying SDP in the early phases of software development is even more beneficial in reducing the cost of quality. In this regard, we expect that the existence of a standard procedure or approach for evaluating and selecting a convenient method for building a prediction model in the early phases could make SDP less challenging, especially for the practitioners. Therefore, in this paper, we identify criteria to evaluate SDP methods and select the best-fit one with respect to stakeholder needs which are elicited via a questionnaire in the early phases of the development lifecycle. We consider stakeholder needs in SDP method selection which is based on multi-criteria decision analysis. We investigate the feasibility of our evaluation approach over an open dataset suitable for early SDP, which initially confirm the feasibility of the proposed approach.