Deep extremely randomized trees

Abstract

In this paper, we propose Deep Extremely Randomized Trees (DET), a deep extension of the extremely randomized trees (Extra-Trees) approach. Our approach unifies classification trees with the layered learning method inspired from deep neural networks. The DET is a deep structure where each layer is a set of Extra-Trees. We look at experimental results on machine learning gold standard datasets and find on-par or superior results when compared to state-of-the-art deep models, with much less parameters to tune. It performs faster than deep neural classifiers, and in most cases even faster than gcForest, without losing accuracy.