Abstract
Choquet integral based on fuzzy measure is a very popular data fusion approach. A major problem in applying the Choquet integral is how to determine a large number of fuzzy measures as the number of attributes increases. The λ-fuzzy measure proposed by Sugeno is a powerful method to resolve this problem. However, the modeling ability of the λ-fuzzy measure is too limited to satisfy actual requirements. In this paper, an extended λ-fuzzy measure is proposed using Shapley value index, and the limitation of the λ-fuzzy measure is significantly overcome under little additional computational loads. The extended fuzzy measure has stronger modeling power than the λ-fuzzy measure, straightforwardly representing interaction among attributes. We apply the extended fuzzy measure to an artificial data set and a real dataset in an iron-steel plant. The results verify the usefulness of the extended fuzzy measure compared with other main existing methods.
* Supported by the National Natural Science Foundation of China (Grant Nos. 60572065, 60532020)
Notes
* Supported by the National Natural Science Foundation of China (Grant Nos. 60572065, 60532020)