Scholarlink

<p>Urban air pollution continues to be a serious environmental and public health issue, especially in Delhi, where the deteriorating Air Quality Index (AQI) has been affected by pollutants including PM₂.₅, PM₁₀, SO₂, and CO. This study leverages time-series forecasting and ensemble machine learning to analyse pollutant dynamics and predict AQI trends. Six models — Support Vector Machine, Random Forest, Linear Regression, Decision Tree, XGBoost, and Artificial Neural Networks (ANN) — are integrated into a proposed method, the Stacked Ensemble Regression Model (SERM), to enhance prediction accuracy. This is evaluated through R², MSE, and RMSE metrics. Feature importance analysis and SHAP (Shapley Additive Explanations) reveal that PM2.5 and CO are the dominant contributors to AQI fluctuations, highlighting the interplay between urbanisation, industrial emissions, and meteorological factors. The framework demonstrates superior performance in short-term pollution forecasting, enabling the development of proactive mitigation strategies to address respiratory health risks. By linking real-time AQI monitoring to climate resilience planning, this research highlights the importance of data-driven decision-making in mitigating community vulnerability to climate-induced air quality hazards. The findings provide actionable insights for policymakers to prioritise emission control measures and advance sustainable urban development goals. Accurate AQI prediction enables urban planners to implement targeted interventions, such as traffic management and emission controls, to improve air quality in sustainable cities and communities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Explainable AI for urban air quality: SHAP interpretation of stacked ensemble AQI forecast

  • 期刊论文

  • 发表日期：

    2025年9月27日

查看全文
  • Gowri L,
  • Nitta Sai Sriya,
  • Muppa Sowmya,
  • Anna Reddy Harshitha Lakshmi,
  • Rengarajan Amirtharajan

摘要

Urban air pollution continues to be a serious environmental and public health issue, especially in Delhi, where the deteriorating Air Quality Index (AQI) has been affected by pollutants including PM₂.₅, PM₁₀, SO₂, and CO. This study leverages time-series forecasting and ensemble machine learning to analyse pollutant dynamics and predict AQI trends. Six models — Support Vector Machine, Random Forest, Linear Regression, Decision Tree, XGBoost, and Artificial Neural Networks (ANN) — are integrated into a proposed method, the Stacked Ensemble Regression Model (SERM), to enhance prediction accuracy. This is evaluated through R², MSE, and RMSE metrics. Feature importance analysis and SHAP (Shapley Additive Explanations) reveal that PM2.5 and CO are the dominant contributors to AQI fluctuations, highlighting the interplay between urbanisation, industrial emissions, and meteorological factors. The framework demonstrates superior performance in short-term pollution forecasting, enabling the development of proactive mitigation strategies to address respiratory health risks. By linking real-time AQI monitoring to climate resilience planning, this research highlights the importance of data-driven decision-making in mitigating community vulnerability to climate-induced air quality hazards. The findings provide actionable insights for policymakers to prioritise emission control measures and advance sustainable urban development goals. Accurate AQI prediction enables urban planners to implement targeted interventions, such as traffic management and emission controls, to improve air quality in sustainable cities and communities.