Can Public Health Policies Drive Carbon Reduction? — Forecasting China's Carbon Emissions and Attribution of Policy Contributions Based on SHAP-BiLSTM
DOI:
https://doi.org/10.65455/wnr6ha05Keywords:
Public Health Policys, Carbon Emission Prediction, Policy Contribution, Machine LearningAbstract
Growing public health imperatives are accelerating carbon mitigation efforts, and carbon emission forecasting constitutes a critical foundation for assessing the feasibility of such measures. However, carbon emission data exhibit pronounced nonlinear and dynamically time-varying characteristics, rendering accurate prediction difficult for traditional econometric models, and existing studies lack quantitative assessment of public health policy contributions. This study proposes the SHAP-BiLSTM analytical approach to ensure both high prediction accuracy and policy interpretability. Based on a daily dataset comprising 2,466 observations spanning January 2019 to September 2025 to conduct quantitative analysis of policy contributions; two public health policy dummy variables and two carbon reduction policy dummy variables are established in chronological order in accordance with IPCC standards. The inclusion of carbon reduction policies helps control for other policy shocks during the implementation period of public health policies. The results indicated that the BiLSTM model significantly outperformed the comparative models in prediction accuracy, demonstrating that excessive algorithmic stacking does not necessarily enhance prediction accuracy. SHAP analysis revealed that public health policies contributed to carbon emission reduction to a certain extent,reflecting that the relevant policies played an effective role in slowing growth during high-emission periods.
References
[1]CHO H, ACKOM E. Artificial intelligence (AI)-driven approach to climate action and sustainable development. Nature Communications, 2025, 16(1). DOI: https://doi.org/10.1038/s41467-024-53956-1
[2]WU K, DING J, LIN J, et al. Big-data empowered traffic signal control could reduce urban carbon emission. Nature Communications, 2025, 16(1): 2013. DOI: https://doi.org/10.1038/s41467-025-56701-4
[3]WANG M. Predicting China's energy consumption and CO2 emissions by employing a novel grey model. Energies, 2024, 17(21): 5256. DOI: https://doi.org/10.3390/en17215256
[4]HUANG H, YI M. Impacts and mechanisms of heterogeneous environmental regulations on carbon emissions: An empirical research based on DID method. Environmental Impact Assessment Review, 2023, 99: 107039. DOI: https://doi.org/10.1016/j.eiar.2023.107039
[5]ZHANG P, GUO Q. How do green municipal bonds affect carbon emissions in China? Evidence from the staggered difference-in-differences approach. Journal of Environmental Management, 2025, 373: 123692. DOI: https://doi.org/10.1016/j.jenvman.2024.123692
[6]XIAO Y, YANG H, XIANG X, et al. Lifting or lessening? Examining the effect of low-carbon city trial on synergistic reduction of pollution and carbon emissions. Energy, 2025, 314: 134195. DOI: https://doi.org/10.1016/j.energy.2024.134195
[7]JIANG H, LIU Z, CHEN Z. The effects of carbon emission rights trading pilot policy on corporate green innovation: Evidence from PSM-DID and policy insights. Sustainability, 2026, 18(3): 1207. DOI: https://doi.org/10.3390/su18031207
[8]JIN Y J, JIN X B, HONG X M, et al. Carbon emission prediction of Hainan Province based on Lasso-Transformer neural network model. Huanjing Kexue/Environmental Science, 2026, 47(2): 781-792.
[9]WANG M, YU J, ZHOU M, et al. Analysis and prediction of carbon emissions from public building operation based on stacking ensemble strategy: The case of Xi'an. Energy, 2026: 139890. DOI: https://doi.org/10.1016/j.energy.2025.139890
[10]WANG Q R, LIU X K, ZHU C F, et al. Carbon emission prediction in China based on LASSO-MEMD-SSA-LSTM. Journal of Safety and Environment, 2026, 26(06): 2475-2484.
[11]YANG Q, JIANG Y H, WU C Y, et al. Carbon peaking prediction for provincial transportation sectors in China based on CNN-LSTM-Attention. Journal of Safety and Environment, 2025, 25(10): 4064-4075.
[12]LI H W, WANG L J, LI S B, et al. Dynamic prediction method for electricity carbon emission intensity based on ISSA-CNN-BiLSTM. Electric Power, 2026,59(7): 66-77.
[13]SIAMI-NAMINI S, TAVAKOLI N, NAMIN A S. The performance of LSTM and BiLSTM in forecasting time series//2019 IEEE International Conference on Big Data. Piscataway, NJ: IEEE, 2019: 3285-3292. DOI: https://doi.org/10.1109/BigData47090.2019.9005997
[14]MOSCA E, SZIGETI F, TRAGIANNI S, et al. SHAP-based explanation methods: A review for NLP interpretability//Proceedings of the 29th International Conference on Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, 2022: 4593-4603.
[15]VAN DEN BROECK G, LYKOV A, SCHLEICH M, et al. On the tractability of SHAP explanations. Journal of Artificial Intelligence Research, 2022, 74: 851-886. DOI: https://doi.org/10.1613/jair.1.13283
Downloads
Published
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
Issue
Section
License
Copyright (c) 2026 The Author(s). Applied Artificial Intelligence Research published by CSTDP
This work is licensed under a Creative Commons Attribution 4.0 International License.
