An analytics framework for healthcare expenditure forecasting with machine learning

The United States healthcare system relies heavily on Medicaid, which serves nearly 80 million people and accounts for a substantial share of both state and federal budgets. This study employs a range of forecasting methods, including ARIMA, Holt's linear trend, polynomial regressions (degree 2 and 4), Prophet, and piecewise linear regression, as well as machine learning models such as random forest, gradient boosting, and support vector regression, to analyze the growth of Medicaid expenditures. Using data from 1966 to 2024, the analysis identifies historical patterns and evaluates model performance with Root Mean Squared Error (RMSE) and related metrics to project costs through 2035. The results show that the autoregressive model with integrated moving average and Prophet generate the most accurate baseline forecasts, suggesting that Medicaid expenditures are likely to exceed one trillion dollars within the next 15 years. Although the machine learning models produced somewhat lower estimates, they revealed complex relationships between policy variables and expenditure behavior, making them useful for building alternative forecasting scenarios. The discussion emphasizes the policy relevance of these findings, particularly in relation to budget sustainability and healthcare equity, and highlights the importance of employing multiple forecasting approaches. Overall, the study demonstrates the value of decision analytics in healthcare forecasting by highlighting the need for accurate predictions, flexible models, and interpretable outcomes. It provides evidence-based tools to anticipate Medicaid's financial challenges and support the development of sustainable healthcare strategies for the years ahead.