Explainable AI for life insurance risk prediction
Ah-ram Lee, Jae Keun Yoo, Jae Youn Ahn · Korean Journal of Applied Statistics · 2025
Predicting the risk level of insurance applicants in life insurance is a critical task for determining premiums and ensuring the stability of the insurance company.Recent studies utilizing machine learning have improved the accuracy of risk prediction.This paper used anonymized applicant information from Prudential available on Kaggle.We employed various machine learning methods, including Random Forest, XGBoost, and logistic regression, as well as deep neural networks, to predict risk and compare their accuracy.We also examined the factors determining the risk level of insurance applicants using the importance of machine learning and explainable artificial intelligence techniques like LIME and SHAP.Among the models used, the deep neural network method yielded the highest accuracy.We identified that the main variables influencing life insurance risk are body mass index, weight, and specific medical history, as well as how these variables determine the direction of risk levels.