Early Software Defect Prediction: Right-Shifting Software Effort Data into a Defect Curve

Predicting the number of defects in software at release is a critical need for quality managers to evaluate the readiness to deliver high-quality software. Even though this is a well-studied subject, it continues to be challenging in large-scale projects. This is particularly so during early stages of the development process when no defect data is available. This paper proposes a novel approach for defect prediction in early stages of development. It utilises a software development and testing plan, and also learns from previous releases of the same project to predict defects. By producing key quality metrics such as percentage residual defects and percentage open defects at delivery, we enable decisions regarding the readiness of a software product for delivery. Over several years, the approach has been successfully applied to large-scale software products, which has helped to evaluate the stability and accuracy of defects predicted at delivery over time.