Search

In the dynamic ensemble selection algorithm, the region of competence of the test sample was composed of fixed samples, which would affect the classifier selection. Therefore, the paper proposed the DES-DCR-CIER algorithm based on dynamic region of competence strategy. Firstly, this algorithm used the heterogeneous classifier to generate the base classifier pool to deal with the problem that the difference between the homogeneous ensemble classifiers was little and the number of heterogeneous ensemble classifiers was small. Next, it applied three steps including the mutual K nearest neighbor with adaptive distance algorithm, approaching the distance center of the sample set and removing the class edge samples to determine the dynamic region of competence of the test sample and it used overall complementarity index to select a set of classifiers. Finally, the classifiers were synthesized by the ER rule to get integration. Experiments on the diagnosis data of breast mass from 8 sonographers in a tertiary hospital in Hefei, Anhui Province and American Wisconsin Breast Cancer Diagnostic data set showed that the diagnostic model based on the DES-DCR-CIER algorithm had better accuracy. [ABSTRACT FROM AUTHOR]