Abstract: This study focused on variable selection using lasso regression to predict future health care costs in claims data. It demonstrated that reducing the number of input features into a linear regression model through variable selection could preserve most of the prediction power. In addition, lasso regression performed only slightly better than linear regression in predicting future costs.
Learning Objective 1: The primary goal of the study is to gain insights into how variable selection from the past history of health care encounters may impact prediction of future health care costs.
Hong Kan (Presenter)
Johns Hopkins University
Hadi Kharrazi, Johns Hopkins University
Hsien-Yen Chang, Johns Hopkins University
Jonathan Weiner, Johns Hopkins University