Machine learning-driven prediction of healthcare resource trends and optimal allocation strategies: a data-driven approach

BACKGROUND: With the advancement of hierarchical diagnosis and treatment systems in China, primary healthcare institutions have become pivotal in delivering basic medical services. Accurate prediction and optimal allocation of healthcare resources are indispensable for improving service quality and ensuring the effective operation of the healthcare system. METHODS: This research utilizes the Sparrow Search Algorithm (SSA) to optimize the hyperparameters of Backpropagation Neural Network (BPNN) and Long Short-Term Memory (LSTM) models, aiming to enhance their predictive performance for primary healthcare resource planning. RESULTS: The findings demonstrate that the SSA-LSTM model significantly outperforms the SSA-BPNN model. Specifically, in predicting the number of primary healthcare institutions, the SSA-LSTM model reduces the root mean squared error (RMSE) by 45% (from 1.429 to 0.78765) and the mean absolute error (MAE) by 36.4% (from 0.99575 to 0.63306) on the test set. Across all prediction tasks, including personnel quantity and total health costs, the SSA-LSTM model achieves an average RMSE reduction of 23.5% and MAE reduction of 17.1% on the test set compared with SSA-BPNN. Similar improvements are evident in the training set, with RMSE and MAE decreasing by 19.2% and 15.4%, respectively. CONCLUSION: The SSA-LSTM model offers robust data-driven decision support for healthcare policymakers. Its superior predictive accuracy enables dynamic adjustments to resource allocation, which is essential for bridging regional disparities in China's primary healthcare system. By accurately forecasting key healthcare indicators, the model facilitates optimized staffing, institutional planning, and budget distribution, thereby laying a solid foundation for evidence-based resource optimization and enhancing the overall efficiency and equity of primary care services.