Learning ELM-Tree from big data based on uncertainty reduction

A challenge in big data classification is the design of highly parallelized learning algorithms. One solution to this problem is applying parallel computation to different components of a learning model. In this paper, we first propose an extreme learning machine tree (ELM-Tree) model based on the heuristics of uncertainty reduction. In the ELM-Tree model, information entropy and ambiguity are used as the uncertainty measures for splitting decision tree (DT) nodes. Besides, in order to resolve the overpartitioning problem in the DT induction, ELMs are embedded as the leaf nodes when the gain ratios of all the available splits are smaller than a given threshold. Then, we apply parallel computation to five components of the ELM-Tree model, which effectively reduces the computational time for big data classification. Experimental studies demonstrate the effectiveness of the proposed method. © 2014 Elsevier B.V. All rights reserved.