A Bayesian network-based scoring methodology for COTS software

There has been an expanding effort by the United States Government to incorporate more commercial off-the-shelf (COTS) software into their complex computer systems. This increased use stems from the realization that pre-existing software products can be a means of lowering development costs, shortening development time, and keeping pace with the changing software market. This increased use has, inadvertently, created a dependence on COTS software that has introduced substantial risks to software procurers, particularly with mission-critical systems. The risks stem from the fact that the source code is generally not available and there is no control over the evolution of the product. Currently, there is no standard "best commercial practice" with regard to the acceptance of COTS software when the source code is not available.; This dissertation presents a Bayesian network-based scoring methodology that aids in determining the acceptability of COTS software. The methodology is premised on the notion that historical data can be used to provide surrogates of probability for COTS software products and the performance of these probabilistic scores can be used to accept or reject a software product. This research introduces a coherent approach for model development and evidence integration where there is no purely objective measure of truth by determining how to structure the COTS software problem, how to develop flexible COTS software requirements, and how to coherently integrate the evidential data to produce probabilistic scores. Moreover, the framework permits evolution of the Bayesian network due to changes in the requirements or the data. A description of the methodology is provided followed by real-world results of the approach applied to the Space Shuttle for real-time operating system (RTOS) selection.