A Hybrid Expert System And Neural Network

A difficult problem of software configuration control is to determine the effects of a contractor's change to "frozen" software. The test team must ascertain the extent and nature of changes and how much regression testing may be needed. A hybrid integration of an artificial neural network (ANN) and knowledge-based expert system (KBE) is proposed to provide parameterized estimates for effects due to software changes. The ANN will be constructed automatically from documentation (i.e. specifications, CASE tools, etc...). Each neuron will represent a module with interconnections representing functional dependencies. The ANN will "learn" cause and effect relationships of software changes by being trained on examples describing actual changes and their effect (failure or non-failure). An operator will perturb the system based on a proposed software change; the KBE will determine partinent information, utilize the ANN to obtain probabilistic estimates of failure, and determine the following parameters: * number of modules indirectly affected * probability a failure will occur within the module(s) * probability a failure will occur in other modules * minimum amount of testing time needed to reach an operator input confidence level * optimal testing strategy given an operator input amount of test time available