CASE-BASED REASONING IN SOFrWARE EFFORT ESTIMATION

A case-based analogical reasoning model, called Estor, was proposed and elaborated from verbal protocols gathered in a prior study. Estor incorporates five analogical problem solving processes: problem representation analog retrieval, solution transfer, attribute mapping, and no-correspondence adjustment. These five generic processes were supplemented with the domain-specific knowledge of the referent expert. The resulting system was then presented with fifteen software effort estimation tasks, ten of which were among those solved by the referent expert, plus five new tasks. For comparison, the expert was asked to estimate the five new tasks as well. The estimates of Estor were then compared to those of the expert as well as those of the Function Point and COCOMO estimations of the projects. Significant between-estimator differences were found, with the human expert and Estor dominating the effects. Correlations between the actual effort values and the estimates of the expert and Estor for all fifteen projects were .98 and .97 respectively. Furthermore, these coefficients differed significantly from those of COCOMO and Function Points. Differences between the model and the referent expert are discussed