Abstract:
|
We discuss ways in which semantic information on software defects can be used in statistical analyses of the software development process. In particular, we describe the orthogonal defect classification (ODC) scheme developed at IBM and discuss a large-scale data analysis project that used defect type and trigger information to cluster a wide variety of development projects into similar groups. The projects within each group were then examined to determine what other attributes were similar, such as product type, code type, process followed, etc. The ability to classify a new project according to these attributes is helpful in determining reliability expectations early in the development process. Additionally, we discuss how ODC information can be incorporated into standard software reliability growth models, using the theory of marked point processes, to give greater insight into dynamics of the software development process during the latter stages. A Bayesian framework allows historical and covariate information to easily be included.
|