EcoSta 2017: Start Registration
View Submission - EcoSta2017
A0712
Title: Is sufficient dimension reduction helpful in prediction? Authors:  Yingcun Xia - National University of Singapore (Singapore) [presenting]
Abstract: I will share some experience of applying the sufficient dimension reduction (SDR) methods to statistical prediction. Using an R package, a number of data sets are studied to check the prediction abilities of models that are constructed based on SDR. We find that for some data sets SDR indeed helps in the prediction, especially when the data structure is complicated, while for most data sets the difference of applying SDR or not is negligible. This observation is basically in line with the theoretical understanding of SDR. Some tips for applying SDR are also suggested.