A value-added predictive defect type distridution model based on project characteristics

Abstract

As software has been pervasive and various software projects have been executed since the 1970``s, software project management has played a significant role in the software industry. There are three major factors in project management; cost, effort, and quality. Much software engineering work has focused on these. Research related to software quality has been focused on modeling residual defects in software in order to estimate software reliability. There are various possible quality characteristics of software, but in practice, quality management frequently revolves around defects, and delivered defect density has become the current de facto industry standard. Currently, software engineering literature still does not have a complete defect prediction for a software product although much work has been performed to predict software quality. On the other hand, the number of defects alone cannot be sufficient information to provide the basis for planning quality assurance activities and assessing them during execution. That is, for project management to be improved, we need to predict other possible information about software quality such as in-process defects, their types, and so on. In this thesis, we propose a new approach for predicting the distribution of defects and their types based on project characteristics in the early phase. For this approach, the model for prediction was established using the curve-fitting method and regression analysis. The maximum likelihood estimation was used in fitting the Weibull probability density function to the actual defect data, and regression analysis was used to identify the relationship between the project characteristics and the Weibull parameters. The research model was validated by using cross-validation methodology.