Design of a new adaptive neuro-fuzzy inference system based on a solution for clustering in a data potential field

In this study, we propose a new method for building adaptive neuro-fuzzy inference systems (ANFIS) via datasets. In order to improve the performance of conventional ANFIS to handle noisy data, we focus on ameliorating the cluster-data space established from a given dataset. To achieve this, we propose a weighted clustering process in the joint input–output data space. Thus, during the clustering process, the cluster with the smallest potential distance, which is a combination of the Euclidean distance and the size of the clusters, has priority when obtaining the surveyed sample. Based on this principle, we formulate a new algorithm for synthesizing an ANFIS via the proposed data potential field, called ANFIS-PF, which has the following features: it establishes a data potential field that covers the entire initial data space, a cluster-data space is built based on the generated data potential field, and the ANFIS is synthesized using this cluster-data space. Finally, we performed experiments using datasets with and without noise to demonstrate the effectiveness of the proposed method in several applications, including dynamic-response noisy datasets obtained from a magnetorheological damper.