In a data set consisting of a large number of attributes, there may be some interaction among the contribution rates from various attributes towards a certain target. Usually, a suitable nonadditive set function (also called a nonadditive measure) defined on the power set of the set of all considered attributes can be adopted to describe such type of interaction and, relatively, a nonlinear integral should be used as an aggregation tool in information fusion. Such a type of nonlinear models has been applying in nonlinear regression and nonlinear classification since the nineties of the last century. In these models, the values of a nonadditive measure are regarded as unknown parameters that can be determined through a learning procedure when a necessary data set is available. However, when the number of attributes is large, the complexity of the computation in the learning procedure is very high and even is unacceptable since it is exponential with respect to the number of considered attributes. So, a compromise strategy, using 2-interactive measure as the nonadditive measure, is necessarily introduced in data mining. It can significantly reduce the computational complexity, though it has to ignore a part of interaction with degrees higher than 2.