Minimum Variance Embedded Random Vector Functional Link Network

Abstract

In this paper, we propose an improved randomized based feed forward neural networks, known as Total variance minimization based random vector functional link network (Total-Var-RVFL) and intraclass variance minimization based random vector functional link network (Class-Var-RVFL). Total-Var-RVFL exploits the training data dispersion by minimizing the total variance while as Class-Var-RVFL minimizes the intraclass variance of the training data. The proposed classification models are evaluated on 18 datasets (UCI datasets). From the experimental analysis, one can see that the proposed classification models show better generalization performance as compared to the given baseline models. © 2020, Springer Nature Switzerland AG.