Construction and application of dynamic classifier based on evidential reasoning rule

Abstract

The classifier based on evidential reasoning rule (ER Rule) provides a new and effective solution for the data classification with uncertainty. The number of referential values for attributes can greatly affect the accuracy and complexity of classification. At present, there are few studies on how to reasonably determine the number of referential values when prior knowledge is scarce. Therefore, a dynamic ER Rule classifier is proposed. Firstly, aiming at minimizing the mean square error, the optimal referential values and weights of the classifier are obtained by intelligent optimization algorithm; Then, the optimal number of referential values is obtained through multiple random experiments, the optimization goal of which is to maximize the accuracy of the test set; On this basis, the implementation of dynamic ER Rule classifier based on PSO is given. Five UCI benchmark datasets are processed to illustrate the effectiveness and stability of the proposed classifier, followed by the identification of algae bloom in remote sensing image, through which the feasibility of the classifier in practical application is verified. The results show that the dynamic ER Rule classifier not only ensures the classification accuracy, but also provides an objective method for determining the number of referential values for attributes, which has a certain reference significance for the popularization and application of ER Rule in other fields.