Extremely Randomized Forest with Hierarchy of Multi-label Classifiers

Hierarchy Of multi-label classifiERs (HOMER) is one of the most popular multi-label classification approaches. However, it is limited in its applicability to large-scale problems due to the high computational complexity when building the hierarchical model. In this paper, we propose a novel approach, called Extremely Randomized Forest with Hierarchy of multi-label classifiers (ERF-H), to effectively construct an ensemble of randomized HOMER trees for multi-label classification. In ERF-H, we randomly chose data samples with replacement from the original dataset for each HOMER tree. We constructed HOMER trees by clustering labels to split each hierarchy of nodes and learns a local multi-label classifier at every node. Extensive experiments show the effectiveness and efficiency of our approach compared to the state-of-the-art multi-label classification methods.