Data-driven hierarchical classifiers based on Choquet integral

The Choquet integral, established with respect to signed fuzzy measure, is an effective aggregation tool in information fusion and classification. Critical coefficients in Classifiers based on Choquet Integral (CCI) are the values of signed fuzzy measure. Currently, determination of these coefficients is either preset subjectively by experience, or retrieved by global optimization methods which are time-consuming, especially when the number of predictive attributes is large. In this paper, an analytic derivation to retrieve the values of signed fuzzy measure in CCI is proposed via discriminant analysis for the first time. On this basis, a generalized Hierarchical Classifiers based on Choquet Integral (HCCI) is established, where a set of scaling parameters is added to CCI to balance the scales of different dimensions. Retrieving of the scaling parameters and the signed fuzzy measure is achieved by a hierarchical structure of program in which a genetic algorithm is embedded with the analytic derivation being proposed in this paper. Performance validation on synthetic and benchmark data sets are conducted to reveal the feasibility and effectiveness of the proposed methods.