Utilizing machine learning to improve healthcare cost prediction on large public datasets

Ensuring good health and well-being is a key United Nations Sustainable Development Goal. However, rising healthcare costs pose challenges, particularly in the U.S., where spending reached $4.9 trillion in 2023, accounting for 17.6% of GDP. This growing financial burden highlights the need for cost prediction models and improved price transparency. This study addresses two key questions: (1) How can computational models predict healthcare costs accurately? (2) How can price transparency be improved? Using New York SPARCS data, and 2 million deidentifed patient records, we developed machine learning models to forecast treatment costs. Our ensemble-based CatBoost approach reduces prediction error (RMSE) by 55% compared to a single CatBoost model. We also optimize prediction coverage to balance accuracy and usability for hospitals and insurers.