The environment effect and health benefit evaluation based on machine learning and prediction

Purpose To maximize health synergistic benefits and provide a scientific foundation for enhancing air pollution control policies, this research proposed a causal inference framework with a decomposition ensemble prediction model for comprehensive policy assessment. Design/methodology/approach We introduced the CD-OASVR-Ensemble-CC method, a policy assessment approach that combined machine learning-based decomposition ensemble prediction (CD-OASVR-Ensemble) with counterfactual contrast (CC) analysis. The CD-OASVR-Ensemble model incorporated support vector regression (SVR) optimized through swarm intelligent optimization algorithms (OA), complete ensemble empirical model decomposition of adaptive noise (CD) and a linear ensemble model (Ensemble). This approach was applied to analyze patterns in historical PM2.5 data and predict the future trend. The predicted PM2.5 served as a counterfactual trend, enabling an evaluation of the policy’s environmental and health effects. Findings Analysis of prediction errors confirmed the proposed CD-OASVR-Ensemble model’s accuracy and robustness in predicting PM2.5 levels. The findings indicated that the implementation of “Regulation 2019” led to a significant reduction in PM2.5 concentration, preventing an estimated 7,960 premature deaths and saving approximately 2,602.95 m yuan in health-related economic costs. Originality/value This study introduces an innovative framework for evaluating the environmental effect of policies, with a specific emphasis on health synergistic benefits.