Final Project for Sta 521 Predictive Modeling at Duke University in 2017 Spring. Finished in team GGGG (Yilin Gao, Wenqi Cheng, Liyu Gong, Mengrui Yin). Credits to my teammates.
The project introduction is here.
The final report is here.
The analysis Rmd file is here. The analysis is done with R.
Simple OLS model, OLS with BIC stepwise variable selection, OLS with BIC and variable transformation, Poisson model, Random Forest, Boosting, LASSO model with interaction terms and variable shrinkage, GAM are explored. Based on RMSEs on both training and testing data, the LASSO model is the best one. Refer to the analysis and writeup files for more details in data processing, model selection, and performance measurement.