A Probabilistic Model for Software Defect Prediction

Abstract

Although a number of approaches have been taken to quality prediction for software, none have achieved widespread applicability. Our aim here is to produce a single model to combine the diverse forms of, often causal, evidence available in software development in a more natural and efficient way than done previously. We use graphical probability models (also known as Bayesian Belief Networks) as the appropriate formalism for representing this evidence. We can use the subjective judgements of experienced project managers to build the probability model and use this model to produce forecasts about the software quality throughout the development life cycle. Moreover, the causal or influence structure of the model more naturally mirrors the real world sequence of events and relations than can be achieved with other formalisms. The paper focuses on the particular model that has been developed for Philips Software Centre (PSC), using expert knowledge from Philips Research Labs. The model is used especially to predict defect rates at various testing and operational phases. To make the model usable by software quality managers we have developed a tool (AID) and have used it to validate the model on 28 diverse projects at PSC. In each of these projects, extensive historical records were available. The results of the validation are encouraging. In most cases the model provides accurate predictions of defect rates even on projects whose size was outside the original scope of the model