Choquet integral agent networks (CHIAN) could realize the operational means embedding existing partial, qualitative knowledge and gray box representations for the information fusion mechanisms. however, the effectiveness of CHIAN for solving real world problems has some limitation, frist, if it his to handle large number of input channels (n) where CHIAN requires (2n ) subset values to determine fuzzy measure (w). second, if human knowledge should be embedded to each input channel of CHIAN agent. from the practical standpoint, these conditions are highly infeasble and are hardly implemented in many realistic problems. thus, this study proposed a method to decompose the input patterns of network structure into some categories. the results showed the effectivenees of the proposed method.