Abstract: Aiming at the problem that the rigor of model selection is not high and the system simplification design is difficult to achieve due to the lack of accurate measurement of the integration priority of the base classifier in the integration process,an integrated method based on sorting selection metrics and adaptive weighting setting was proposed.Firstly,the K-fold cross-validation and the combined index metric method constructed by combining the error entropy of the design and the complementarity of the classifier are utilized to select two classi-fiers with the highest integration prio-rity.Then,considering the integration influence between the remaining candidate classifiers and the selected classifier subsets,the overall combination index metric based on combination index is constructed to realize the prioritization of different models.Finally,the best weights are found for different models for integration classification by adaptive weight method.The experimental results on the UCI dataset show that compared with other classification models,the classification evaluation indicators of the proposed method are improved,proving the feasibility of the integration method.This method selects quantitative basis of design model and adaptive weight setting mechanism through the designed model,making the whole integrated classification system have the stratification for model selection and the characteristics ofadaptivesimplification system.

Key words: Adaptive weight, Combination index, Complementarity, Error entropy, Sorting selection