An aggregation approach to the classification problem using multiple prediction experts

A general problem of combining the classifications or predictions of n loca l information systems S-1, S-2, ..., S-n into a single system S is consider ed in this paper. We introduce a new approach for solving an aggregation pr oblem where a decision maker needs to classify an observation based on grou p-membership predictions coming from multiple experts. We empirically test our approach along with two alternative approaches found in the literature using real-world data and four popular classification techniques. The resul ts show that our approach produced the best performance according to certai n performance measures. Researchers may use this method to create a knowled ge base of information for a distributed expert system or one that acquires prediction information from the Internet. (C) 2000 Elsevier Science Ltd. A ll rights reserved.