XGBoost in Insurance 2017

Data and Code to reproduce results for my talk at Paris: R in Insurance 2017 Conference

Non life pricing: empirical comparison of classical GLM with tree based Gradient Boosted Models

To reproduce the results, please:

• clone the repository
• set your working directory to the cloned depository
• Inside the code folder run:
  • 00_init_* to initiate the workspace correctly.
  • 02_train_* to execute parameter tuning and save xgboost models.
  • 03_scoring to execute cross-validation and obtain results, on both GAM and xgboost.

Further, 01_preprocess_*.R code to prepare the dataset is provided. Note: xgboost mdoels are not provided directly since they exceed the size limit of GitHub. Feel free to reach out and I will provide them privately if pre-tuning is too expensive.

Thank you!