Research on Prediction Methods of Anthropogenic Economic Pollutant Emission Index by Coupling of Ensemble Machine Learning and Time-Series Models Under Multiple Features
Year: 2026
Type: article
Abstract: Predicting the anthropogenic economic pollutant emissions index helps balance economic growth and environmental protection. In this study, a set of constructed features was derived using provincial-level basic data from 1995 to 2023. By constructing a weighted ensemble strategy incorporating Extreme Gradient Boosting, light gradient boosting machine, Random Forest, and Multi-Layer Perceptron, and integrating it with the Autoregressive Integrated Moving Average model and Shapley Additive Explanations, an anthropogenic economic pollutant emission index (AS_GPI) forecasting model was finally established, with basic and constructed features employed as its inputs (AFEA-AG). The AS_GPI forecasting model (EA-AG) was developed without constructed features, with all other settings consistent with the AFEA-AG model for comparison. Results show that the proposed model achieves high forecasting accuracy for AS_GPI across four typical pollutants, with R2 values exceeding 0.95 for the AS_GPI_NOx and AS_GPI_PM2.5. The mean absolute percentage error was as low as 0.0744. The forecasting model with constructed features yielded lower errors and higher stability in its prediction results compared with the one without such features. Feature contribution analysis revealed differing key contributors, with AS_GPI lagged values and economic-related characteristics among underlying variables exhibiting strong predictive importance. The 2024 projection results indicated certain disparities in pollution control effectiveness between key and non-key regions. Further analysis of historical and predicted data revealed a nationwide decline in AS_GPI between 1995 and 2024. Beijing and Shanghai achieved notable environmental quality improvements through anthropogenic emission reductions. The coefficient of variation values of different AS_GPIs reveal spatial heterogeneity and differences. Nationwide efforts should prioritize the control of anthropogenic NMVOC and NOx emissions. This framework provides a prediction method that offers certain reference for the development of the economy and the environment.
Source: Atmosphere
Authors: Mengzhen Li, Yang Cao, Jianlei Lang
Institutions: Beijing University of Technology, Beijing Wuzi University
Language: English
Subfield: Economics and Econometrics
Domain: Social Sciences
SDG: Climate action
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7160599650