Graph embedded intuitionistic fuzzy weighted random vector functional link network

Abstract

Random vector function link (RVFL) network with direct links and closed form solution has shown its strength among randomized neural networks (RNNs). RVFL assumes that all the samples are equally important, however, this may not be satisfied in the real world scenario. Moreover, RVFL ignores the geometric and discriminative information of the data. To overcome these shortcomings, we propose a graph embedded intuitionistic fuzzy weighted random vector functional link (GE-IFWRVFL) network. The proposed GE-IFWRVFL model assigns weights to each data point based on the membership and non membership functions. Membership values are assigned based on the distance of each sample from the class centre and non-membership values are assigned based on the sample distance as well as the heterogeneity of the samples. Also, the proposed GE-IFWRVFL exploits the geometric data relationship of the data under the Graph Embedding (GE) framework to enhance the generalization performance. A novel regularization term is incorporated in the proposed GE-IFWRVFL model to handle the topological structure of the data. Similar to intuitionistic fuzzy RVFL (IFRVFL) model, the proposed GE-IFWRVFL optimizes the output layer weights assigning appropriate weights to each sample. The performance of the proposed GE-IFWRVFL model is evaluated over 25 UCI and KEEL datasets. Experimental results demonstrate the better generalization performance of the proposed model compared to the baseline models. © 2022 IEEE.