Title: Applying neural networks to predict government funding needs, taking subsidies for low-income elderly living allowances as an example

Authors: Yi-Hao Chen; Wen-Chen Huang; Hsiu-Pi Lin

Addresses: National Kaohsiung University of Science and Technology, 2 Juoyue Road, Nantsu, Kaohsiung, 811, Taiwan ' National Kaohsiung University of Science and Technology, 2 Juoyue Road, Nantsu, Kaohsiung, 811, Taiwan ' Chia-Nan University of Pharmacy and Science, No. 60, Sec. 1, Erren Rd., Rende Dist., Tainan City 717301, Taiwan

Abstract: This study develops and evaluates machine learning models for forecasting social welfare budgets, focusing on the 'living allowance for low-income senior citizens' in Taiwan. We compare neural network models against traditional forecasting methods to assess their potential for improving budget prediction accuracy and efficiency. Our experiments utilise dense neural networks (DenseNN) and recurrent neural networks (RNN) with non-time and time-series data. Results demonstrate that the best-performing neural network model outperforms traditional growth rate forecasting methods in accuracy. The study highlights the impact of political decisions and price index changes on statistical data, emphasising the need for adaptive forecasting models. Our findings suggest that machine learning approaches can enhance budget forecasting accuracy in social welfare programs, potentially improving resource allocation and policy planning.

Keywords: elderly subsidy; administrative statistics; machine learning; ML; neural network.

DOI: 10.1504/EG.2025.149237

Electronic Government, an International Journal, 2025 Vol.21 No.6, pp.692 - 719

Accepted: 28 Oct 2024
Published online: 20 Oct 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article