Prediction of Health Care Insurance Expenses using Machine Learning and Artificial Neural Network

The application of artificial intelligence (AI) in health insurance is an important innovation that is rapidly driving improvements and redefining the interaction between the insurer and policyholder. It contributes to improving and streamlining the user experience, as well as making operations easier. Machine learning and AI can help insurers more correctly analyze the risks involved in underwriting a policy, and thus estimate premiums and losses. Traditional loss models in healthcare, such as Generalized Linear Models and Multiple Linear Regression, rely on statistical methods and actuarial assumptions, whereas machine learning models leverage data-driven algorithms and patterns to predict healthcare expenses with greater accuracy, adaptability, and computational speed. The purpose of this study is to estimate and predict health insurance expenses based on several variables including age, BMI, sex, number of children, region, and whether the individual is a smoker. The prediction is conducted using several machine learning algorithms, including Regression Decision Tree, Gradient Boosting Machine, XGBoost, Multiple Linear Regression, and Feedforward Artificial Neural Network. Since the dataset used in the study has a highly skewed distribution for expenses, a logarithmic transformation is applied on the data of expenses, to be used in predicting the logarithm of expenses to improve each model’s accuracy. The results show that the Feedforward Artificial Neural Network performed the best, with mean absolute percentage error of 2.11% to predict the logarithm of expenses.