Class Binarization to NeuroEvolution for Multiclass Classification
Authors: Gongjin Lan, Zhenyu Gao, Lingyao Tong, Ting Liu
Summary: Multiclass classification is a elementary and difficult activity in machine studying. The present strategies of multiclass classification could be categorized as (i) decomposition into binary (ii) extension from binary and (iii) hierarchical classification. Decomposing multiclass classification right into a set of binary classifications that may be effectively solved by utilizing binary classifiers, known as class binarization, which is a well-liked method for multiclass classification. Neuroevolution, a basic and highly effective method for evolving the construction and weights of neural networks, has been efficiently utilized to binary classification. On this paper, we apply class binarization strategies to a neuroevolution algorithm, NeuroEvolution of Augmenting Topologies (NEAT), that’s used to generate neural networks for multiclass classification. We suggest a brand new methodology that applies Error-Correcting Output Codes (ECOC) to design the category binarization methods on the neuroevolution for multiclass classification. The ECOC methods are in contrast with the category binarization methods of One-vs-One and One-vs-All on three well-known datasets Digit, Satellite tv for pc, and Ecoli. We analyse their efficiency from 4 points of multiclass classification degradation, accuracy, evolutionary effectivity, and robustness. The outcomes present that the NEAT with ECOC performs excessive accuracy with low variance. Particularly, it exhibits vital advantages in a versatile variety of binary classifiers and powerful robustness.
