Imputation of the algorithms for certainty factor manipulation by individuals using neural networks and regression: a comparison to expert system shells

Imputing functions that emulate individuals´ manipulations of certainty factors are discussed. A comparison of regression-based models, neural network models, and commercial expert-system shells is presented. It is established that both regression and neural models built on half the data are statistically better predictors than the algorithms embedded in the shells. Because of multivalued responses, neither type of model can fit the data exactly. A strawman target model which averages multivalued responses to obtain a single-valued response is built. The regression and neural models are found not to be statistically different from each other; both types of models are found to be statistically inferior to the strawman model and superior to all but one of the shells. The implications are discussed and directions for further research are identified