Precise determination of non-additive measures

The Choquet integral model has been shown useful in many practical applications due to its distinguished feature that the interaction among predictive attributes toward the objective attribute can be properly reflected through a set of non-additive measures. Non-additive measures are traditionally determined using heuristic strategies or genetic algorithms; however, their solutions to Choquet integrals are often not unique or deterministic. Besides, raw attribute data are often not good enough to identify all coefficients, which leads to an incomplete Choquet model. In this paper, we present an efficient approach to determine the complete non-additive measures. The approach adopts a new data normalization, based on median alignment, for the Choquet model to overcome the above practical deficiencies raised in traditional methods. We show how the Choquet model is applied to the significance analysis for wireless cross-layer network design.