Optimizing Government Funding Allocation with Ensemble Machine Learning Models

Given the limited availability of state funds, managing state finances in an effective, efficient, and prudent manner is essential. High demand for additional state funds without sound justification can lead to inefficiencies and poor budget execution. Therefore, decisions regarding the allocation of additional funds should be made selectively and based on historical data reflecting the quality of the work unit in budget management. This research aims to develop a machine learning application to assist decision-makers in recommending additional funding percentages. Utilizing the Budget Implementation Quality Indicator (IKPA) data, we performed feature selection using Principal Component Analysis (PCA), resulting in three selected features. These features were then used to build models with base models (Decision Tree Regression, K-Nearest Neighbor Regression) and ensemble learning methods (Stacking, Bagging, Random Forest, Boosting: AdaBoost, XGBoost, LightGBM). After to-Fold Cross-Validation and hyperparameter tuning, LightGBM demonstrated the lowest error rate with a Root Mean Square Error (RMSE) of 0.0646 and a Mean Absolute Error (MAE) of 0.0520, outperforming XGBoost in predicting additional fund allocation proportions. The application supports informed and accountable financial decision-making, promoting efficient and prudent national financial management. By comparing base models and ensemble techniques, this research provides critical insights into machine learning applications in financial management, driving methodological innovation and advancing the field.