Defect prediction using historical defect density can provide an early look into inadequate testing resources, ineffective test cases, improper code coverage, and sub-standard code quality. Better understanding of these indicators helps a development team focus on creating more effective test cases and test procedures, introducing code reviews, and balancing resources to get a better quality product. The author has utilized the historical defect density and defect partitioning to predict total number of defects on two projects. Although the success has been limited due to the lack of historic data, the technique has compelled to search answers to questions including: are we finding enough defects and why are we seeing so many defects at such an early stage. By assessing the number of lines to be written for a product and using a shop's historic defect density, one can predict the total number of defects. Based upon an organization's past experiences, the prediction can be fitted to a pattern that typically represents defect arrival rate in calendar time during a product development. The prediction depends upon accuracy of estimated lines of code and the similarity of the development environment for which the historical density was calculated to the current environment. It is imperative that the counting mechanism for all systems be the same. If it is not, the estimated lines of code will differ substantially and the total estimated defects will be off by a wide margin.
展开▼