An Extreme Learning Machine Method for Multi-Classification with Mahalanobis Distance
Shujuan Zhou, Qiang Wang, Yilin Fang, Quan Liu · 2018
I23 this paper, a novel approach for multi-classification problem, named decomposed extreme learning machine binary tree (DELM-BT), is proposed. The DELM-BT is an ensemble of ELMs that work in a decomposition mode. Each binary classification ELM are combined in a binary tree (BT). During the structural optimization of BT, Mahalanobis distance (MD) is taken into consideration as the inter-class distance due to the inter-class association consideration of MD, and Euclidean distance (ED) is used to measure intra-class distance. At last, the vibration signals of gear are utilized as the illustrated examples. The experimental result demonstrates that the time consuming of the proposed method is less than other existing classifiers under the same classification accuracy.