A new fuzzy cognitive maps classifier based on capsule network

Abstract

In recent years, classification based on fuzzy cognitive maps (FCM) has received extensive attention, has been applied in many engineering problems, and has proven to work well. However, a host of the existing methods are designed for specific engineering problems, which lack universality and low classification accuracy. In this study, a straightforward and versatile model is proposed to deal with classification. The key of the model is to integrate the capsule network into the inference rules. It forms a new inference rule with a strong coupling coefficient. Weights were learned with a particle swarm algorithm, and the cross-entropy with constraints was used as the cost function. The goal of this research is to maintain the interpretability of the model and enhance its universality and classification performance. The top-down feedback mechanism of the capsule network can meet this requirement. A series of experiments involving public data proved that the proposed method achieves better performance than the previous results obtained using low-level cognitive maps. The comparative analysis also demonstrates the performance of the classification delivered by the proposed model.