Archives

Hyperparameter Optimization in XG Boost for Insurance Claim Prediction


I. Gede Pajar Bahari and Hendri Murfi
Abstract

Machine Learning methods are beneficial for solving various problems, especially big data. One issue related to the big data is the prediction of insurance claims in the insurance industry. The XGBoost is a machine learning method using ensemble learning with decision trees as its base model. XGBoost consists of hyperparameters that need to determine before the training process. The partial grid search is a hyperparameter optimization usually use for XGBoost. In this paper, we apply and analyze another optimization method to XGBoost called Bayesian search for two problems of the claim prediction, i.e., regression and classification. Our simulations show that the partial grid search gives slightly better accuracy compared to the Bayesian search. However, the Bayesian search provides a significantly faster running time than one of the partial grid search.

Volume 12 | 04-Special Issue

Pages: 1510-1517

DOI: 10.5373/JARDCS/V12SP4/20201630