Forecasting financial support for government institutions using artificial intelligence: The case study of museums

Subject. Forecasting the financial support of public institutions based on regional panel data of the constituent entities of the Russian Federation. Objectives. To develop and test a methodology for forecasting the financial support of museums using artificial intelligence methods. Methods. A panel dataset of 85 constituent entities of the Russian Federation for 2018–2024 was compiled. Machine learning methods were applied, including gradient boosting, regularized linear models, and their ensembles. Hyperparameter tuning was carried out using an artificial intelligence agent (AI agent) based on the Qwen3.5‑27B language model. Thirty‑two computational experiments were conducted. Forecast accuracy was assessed using the median absolute percentage error (MdAPE); rolling validation was applied over four test years (2021–2024). Results. The gradient boosting model achieves an MdAPE of 8.1% across regions, outperforming the best naive forecast by 1.9 percentage points and the previous‑year forecast by 4.6 percentage points. The average MdAPE under rolling validation was 12%. For 64% of the regions, the forecast error does not exceed 15%. Infrastructure‑related characteristics of institutions (28.6%) and personnel‑related characteristics (22%) demonstrate the highest predictive power. A systematic tendency to underestimate forecasts for large regions was identified; this is driven by discretionary budgetary decisions that cannot be predicted based on operational indicators. Conclusions and Relevance. Artificial intelligence methods demonstrate an advantage over traditional econometric approaches in the context of nonlinear relationships and interregional heterogeneity. The proposed methodology is applicable to forecasting financial indicators of public institutions in other sectors, provided that similar operational data at the regional level are available. The results can be used by public authorities in budget expenditure planning and in preliminary assessments of expected levels of financial support.