Analysing Software Metrics for Accurate Dynamic Defect Prediction Models

Abstract

Defect Prediction Models are widely used in many software products. These estimates help assessing the risks of defect leakage objectively. However, Most of the published Defect prediction models do not necessarily fit in providing estimates for large complex telecom billing solutions. We have tried to analyze the accuracy of the Prediction model based on various software metrics. It is important to understand the relation between different software metrics for accurate software defect prediction. The Telecom Billing solutions have complex applications, configurations, size and parameters. We have tried to correctly identify the Software Metrics that affect the accuracy of the software Defect prediction models. This paper presents an evaluation; software metrics are investigated in order to identify the ones which significantly impact the accuracy of defects prediction models. We found that Historical Defect Density, Removal rates can help in accurate Defect prediction. Data from various projects is collected and analyzed. The prediction of the models, with and without metrics is compared. To access whether the improvements are significant or not, analysis on the accuracy of the Defect prediction is done. It was found that metrics such as Defect Density, Removal rates, trend of absolute number of defects along with Effort for a particular application, significantly affects the accuracy of a Defect prediction tool. Also behavior of one application may or may not be similar to another application. Thus, to have an accurate prediction, we should attempt at avoiding generalization.