A multi-source transfer-based decision-making method with domain consistency and contributions

Large volumes of historical data that characterize the preferences of decision-makers are generated and recorded. Each decision-maker's historical data contain different knowledge and reflect different preferences. However, the learned decision models based on data-driven multi-criteria decision-making (D2MCDM) cannot help generate satisfactory decision solutions when the available data are insufficient. To solve this problem, the parameterbased transfer learning with multi-source domains (multi-SDs) is considered to be integrated with D2MCDM. Following the idea, this paper proposes a multi-source transfer-based decision-making (MTBDM) method with domain consistency and contributions to acquire the preferences of decision makers. The stability of criterion weights and that of decision makers' individual consistency are constructed and combined with the difference between data distributions of multi-SDs and the target domain (TD) to develop the domain consistency. Based on the developed domain consistency, two optimization models are constructed to transfer criterion weights and individual consistency between multi-SDs and the TD, in which domain contributions of multi-SDs to the TD are considered based on domain consistency to avoid negative transfer. The advantages of the MTBDM method are demonstrated by its application to the auxiliary diagnosis of breast lesions. Its predictive performance is further validated by its comparison with the MTBDM method in four different situations and existing methods.