Abstract
Fuzzy preference based DS evidence theory effectively solves the combined explosion ascribing to the increasing of the cardinality of the frame of discernment in the application. However, limited work has focused on evaluating the performance of the binary classifier (i.e. dynamic improvement of the according fuzzy preference element). The conflict coefficient can well reflect the relationship between various pieces of evidence and effectively evaluate the ability of each classifier. By statistically analyzing prior conflict information, the conflict interval range is determined to more effectively correct the fuzzy preference matrix. Therefore, we proposed a new method to evaluate the competence of fuzzy preference using statistical conflict in the frame of DS evidence theory. First, a binary transformation of the multiclassification problem is performed to generate the BPA and the matrix was generated by orthogonal fusion of multiple BPAs. Second, calculate the conflict coefficients and make statistics. The statistical conflict coefficients are combined and the conflict interval is solved to modify the construction of the fuzzy preference matrix. Third, based on the measured data in the feature space, the matrix is constructed by majority voting using the k-nearest neighbors of correctly classified objects. Finally, the two matrices are fused by determining the relationship between the conflict coefficient and the threshold value of each binary pair. The non-dominance degree is calculated based on the modified fuzzy preference matrix to make the final decision. Analysis and results of application examples and several datasets in practical application are used to demonstrate effectiveness and flexibility of the proposed method.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call