Regression-Based Machine Learning Models to Predict Healthcare Costs

Abstract— Accurately predicting healthcare costs is crucial for insurance companies and governments. Machine learning has become increasingly popular in providing more precise methods for predicting healthcare costs. In our research, we conducted several experiments to develop regression models based on healthcare cost prediction. We processed public data from Kaggle and prepared it to train various regression models, including Linear Regression (LR), Random Forest Regressor (RFR), and Gradient Boosting (GB). After several training trials, we evaluated the models' performance using multiple methods such as R-squared, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE), along with learning curves, to compare their efficiency and accuracy. We then selected the most optimal model, which was RFR, optimized using GridSearchCV, and achieved a promising result , R-squared score of 87.17% in predicting healthcare costs.