Résumé: Given a specific decision model for two decision problems faced by a decision maker, decision parameters can be learned from the accumulated historical data. In general, more accumulated data can better reflect the real preferences of the decision maker. The knowledge contained in more historical data for one decision problem may help improve parameter learning for the other decision problem with less historical data. Inspired by this idea, a cross-domain decision making method is proposed to improve the learning of parameters, namely criterion weights and risk attitudes of the decision maker, in the target decision problem with less historical data by using the parameters learned from more historical data in the source decision problem. A transferability measure is developed to evaluate whether the knowledge contained in the historical data from the source decision problem is beneficial for improving parameter learning in the target decision problem. When the transferability between the source and target decision problems is judged to work well, the parameter transfer strategy is designed to conduct the parameter learning in the target decision problem by using criterion weights and risk attitudes of the decision maker learned from historical data from the source decision problem. The effectiveness of the proposed method is validated by its application in helping diagnose breast lesions with the historical diagnostic data of seven radiologists collected from a tertiary hospital located in Hefei, Anhui, China.

Langue: Anglais