A Machine Learning and Ratemaking Evaluation of Four Auto Insurance Pure Premium Modeling Algorithms

Actuarial science is increasingly using machine learning. Machine learning has improved actuarial pricing prediction by forecasting future claims. Insurers are testing these algorithms, but model explainability and implementation cost are issues. The current literature barely touches upon this broad field of inquiry. Fujita et al. published a 2020 study on predictive frequency models using AGLM, GLM, GAM, and GBM. König and Loser (2020) compared GLMs, neural networks, and XGBoost for frequency prediction. We want to add to existing research by forecasting pure premium rather than frequency or severity, comparing four methodologies, and analyzing and presenting a wide range of quantitative and qualitative performance evaluation metrics. This paper compares four approaches for predicting pure premium: GLM, AGLM, XGBoost, and neural networks. On a test set, the study compares each model's pros and cons in P&C insurance pricing, as well as its quantitative and qualitative performance. Despite these obstacles, machine learning in actuarial pricing has clear benefits. By may improve risk profiles, pricing accuracy, profitability, and market competitiveness by using massive volumes of data and advanced analytics. However, getting these results needs a holistic approach that balances machine learning's technical sophistication with insurance's regulatory and operational limits. Further research should construct interpretable machine learning models customized to insurance pricing needs and limits while addressing the ethical, regulatory, and implementation obstacles to their adoption.