Software Defect Prediction Modeling

Defect predictors are helpful tools for project managers and developers. Accurate predictors may help reducing test times and guide developers for implementing higher quality codes. We propose a statistical defect predictor model with two major differences from the existing ones. Our model will use static code measures as input, since they can easily be collected with automated tools and prevent any kind of human subjectivity. There are two major differences between our proposed model and the existing ones. While existing defect predictors treat the software metrics equally, we claim that the effect of different metrics on prediction performance may vary. Thus, we propose a weighted prediction model where the weights are estimated according to the relative importance of metrics. Furthermore, our proposed model assumes that the software metrics are not statistically independent from each other and incorporates the relations between metrics into prediction process. Finally we propose a framework that can use existing set of software metrics and produce more accurate estimates of module complexities by using software system structure information.