Trade-Offs between Early Software Defect Prediction versus Prediction Accuracy

Abstract

In any software development organization, reliability is crucial. Defect prediction is key in providing management with the tools for release planning. To predict defects we ask the question of how much data is required to make usable predictions? When testing, a rule of thumb is to start defect prediction after 60% of system test has been accomplished. In an operational phase, managers cannot usually determine what constitutes 60% of a release and might not want to wait that long to start defect prediction. Here we discuss the trade-offs between the need of early predictions versus making more accurate predictions.