Aggregating and updating experts' knowledge: An experimental evaluation of five classification techniques

摘要

Knowledge acquisition consists of eliciting expertise from one or more experts in order to construct a knowledge base. When knowledge is elicited from multiple experts, it is necessary to combine the multiple sources of expertise in order to arrive at a single knowledge base. In this paper, we present and compare five techniques for aggregating expertise. An experiment was conducted to extract expert judgments on new product entry timing. The elicited knowledge was aggregated using classical statistical methods (logit regression and discriminant analysis), the ID3 pattern classification method, the k-NN (Nearest Neighbor) technique, and neural networks. The neural net method was shown to outperform the other methods in robustness and predictive accuracy. In addition, the explanation capability of the neural net was investigated. The contributions of the input variables to the change in the output variable were interpreted by analyzing the connection strengths of the neural net when the net stabilized. We conclude by discussing the use of neural nets in knowledge aggregation and decision support.