A Comparative Study of Regression Methods for Predicting Family Health Insurance Expenses

Accurate prediction of family health insurance expenses is essential for insurers and policyholders to optimize premium pricing and financial planning. This study provides a comparative analysis of various regression techniques, including Multiple Linear Regression, Ridge Regression, Lasso Regression, Support Vector Regression (SVR), Gradient Boosting Regression, Decision Tree Regression, and Random Forest, using a benchmark data set that contains demographic, socioeconomic and health-related variables. We assess the models’ performance in terms of accuracy, interpretability, and robustness, focusing on challenges such as multicollinearity, outliers, and nonlinearity. The findings indicate that while traditional linear models establish a baseline, machine learning approaches, especially those utilizing regularization, significantly enhance prediction accuracy. Notably, Random Forest achieved better performance across multiple metrics, including residual sum of square, and, R-squared, and Adjusted R-squared. This motivated the use of Polynomial Regression, which, after fine-tuning, outperformed all the other models. Our research contributes to developing more accurate predictive models, supporting insurers in risk assessment and premium pricing strategies.