Abstract
In this article, we review main tree-structured approaches with their recent advances. CART methodology is fundamental and unavoidable in understanding and developing a variety of treestructured approaches. In particular, MARS method provides a useful expansion to the continuous model of CART. In recent advances, some methods developed via combining with ensemble learning method obtain more powerful prediction performance and provide more attractive variable importance. For applied researchers, we present the illustration of applications and a small scale simulation in main tree-structured approaches. Finally, we summarize characteristics of methods and provide some works of interest in the future.
