Evaluating the acceptance of an insurance policy requires the underwriter's accuracy in assessing prospective customers' risk profiles. If it is discovered that a prospective customer has a high risk, the policy application will automatically be rejected. It will take time and substantial costs if someone relies on the traditional underwriting process. Therefore, this research tries to apply the boosting machine learning method in classifying the risk level of life insurance policies using data from Prudential Life Insurance Assessment 2015. There are four boosting techniques used, namely multi-class adaptive boosting (AdaBoost), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and categorical boosting (CatBoost). XGBoost and LightGBM perform best when training through hyperparameter optimization and experimenting on test data; both have a weighted F1 score of 0.48, with LightGBM performing much faster.