Splitting along Directions: an application to the study of the U.S. business cycle

The talk deals with the problem of dimension reduction in the general context of supervised statistical learning. In particular, data mining applications with many covariates often lead to high dimensional data analysis problems. The main goal of the proposed methodology is to improve tree based methods as prediction tool by introducing an alternative approach to data partitioning which is meant to handle large numbers of (correlated) covariates. The key idea is to use suitable combinations of covariates recursively identified. This modified recursive partitioning is adopted to investigate the structure of expansions and recessions on U.S. business cycle