A Comprehensive Evaluation of Machine Learning Vs. Statistical Learning for Medical Insurance Premium Prediction

This research paper compares different machine learning (ML) algorithms and statistical learning models used to predict medical insurance premiums. Medical expense forecasting can help insurance providers and policyholders take informed decisions about health care services and premiums. The various models of statistical learning were compared with the ML algorithms: Random Forest, Regression Trees, and Support Vector Machine. The models were tested on an insurance premium data set, which included a number of predictive factors such as geolocation, smoking habits, number of children, Body Mass Index (BMI), age and gender. The performance of the model was evaluated by mean absolute error (MAE), root mean square error (RMSE), and R-squared (R²). The performance of PLS was the best among the statistical regression methods, as revealed by the results, with the least MAE, RMSE and highest R² values than classical regression and Elastic Net Regression. In general, the ML algorithms performed better than the statistical learning models, with the best overall predictive performance obtained by the Random Forest algorithm. The results indicate the application of ML algorithms can be used to predict future healthcare costs and to help insurance companies when determining insurance premiums. Moreover, when the classical regression assumptions (e.g., the normality of data) are not easy to meet, especially in the presence of available statistical software like R, ML methods offer useful alternatives to modelling.