Identifying metrics for commercial-off-the-shelf software with inductive inference based on characteristic vectors

Nowadays, many users and organizations are interested in acquiring COTS (commercial-off-the-shelf) software products instead of building software systems themselves as acquisition reduces development costs. COTS products are usually provided in a packaged style without the source code but with many ready-to-use functions. To assure the proper level of quality, many organizations provide quality evaluation and certification services for COTS. Generally, their vendors are reluctant to disclose the source code. Thus, the major way of quality evaluation and certification requires dynamic behavior testing, essentially black-box testing. Since observing every aspect of external software behavior is almost impossible, it is crucial to designate an adequate range for quality evaluation such as an adequate number of quality checklists or product quality metrics for external behavior testing. Hence, to establish rules of selecting quality evaluation criteria in systematic ways, there have been attempts to analyze and utilize the past records of software evaluation based on artificial intelligence techniques. A Bayesian belief network (BBN) is one of the methods using an inductive inference based on prior experiences. In this paper, we represent software as characteristic vectors having dependency relationships with the external product quality metrics. BBN is then used to infer the metrics for new software products.