Neighborhood rough cognitive networks

Abstract

Rough cognitive network (RCN) is a granularity neural network combining fuzzy cognitive maps with rough set theory, which has the advantages of transparency and interpretability. However, When the existing RCN are applied to the decision-making aspects of practical problems, they exhibit problems such as weak generalization and unstable performance. To solve the above problems, a highly efficient classification decision model neighborhood rough cognitive networks (NRCN) is proposed. It uses a feature selection algorithm based on the neighborhood rough set to minimize the data dimensionality. Moreover, the introductions of sub-neurons and Hellinger distance in proposed model can better discover knowledge and make more reasonable inference decisions. Various open datasets and real financial data are implemented to verify the effectiveness and validity of the proposed method.