Research is proposed to investigate the extent to which the role of expert panels in assigning priority to Aegis software change requests can be partially or completely replaced by a consistent, objective, and cogent automated machine-learning system. Three major aspects of each change request are to be assessed separately, 1) complexity, 2) operational impact, and 3) mission impact. Each of these will be assessed using three complementary quantitative models: linear, neural, and case-based reasoning. The resulting nine models will be trained using a relational database representation of past change requests, created using text mining. The approach under study uses an optimized ensemble learning function combining linear and quantile information into separate aggregate ratings for complexity, operational impact, and mission impact before aggregating those three aspects into an overall numeric assessment for priority. Each assessment will have a machine-generated explanation and a quantitative certainty measure.
Benefit: This approach provides an unbiased evaluation of the Change Request using a panel of synthetic experts. The
Keywords: Neural Nets, Neural Nets, , case-based reasoning. , Linear models, panel of experts, ensemble learning, Artificial Intelligence, Change Management,
