Abstract:
|
In this poster, we propose a new tree-structured regression method using a projection pursuit approach. It extends the projection pursuit classification tree to the regression problem. The main advantage of the projection pursuit regression tree is the exploration of the independent variable space in each range of the dependent variable. Also, it keeps all the main properties of the projection pursuit classification tree. To improve the predictability, the projection pursuit regression tree provides several methods to assign values in the final node. With this development, we can easily explore the data space for the piecewise regression and find a better model with good predictability.
|