Scholarlink

<p>Air is a vital natural resource that sustains life by providing oxygen and maintaining the balance of the Earth’s ecosystems. Clean air supports human health, climate regulation, agricultural productivity, urban planning, and overall well-being. However, increasing concentrations of harmful gases and particulate matter released through anthropogenic and natural activities have significantly degraded air quality in many regions worldwide. Accurate forecasting of air pollution levels has therefore become essential for effective environmental monitoring and public health protection. In recent years, Artificial Intelligence (AI)-driven approaches have gained prominence for modeling the complex and non-linear dynamics of air pollution. The article presents a comprehensive literature survey on air quality index (AQI) prediction using standalone deep learning (DL) models, hybrid DL architectures, and federated learning (FL) frameworks. The survey systematically reviews pollutant characteristics, meteorological factors influencing air quality, commonly used datasets, data preprocessing strategies, and performance evaluation metrics reported in recent studies. Further, this review article highlights key methodological challenges, comparative performance trends, and emerging research directions, with particular emphasis on privacy-preserving and decentralized learning paradigms enabled by FL. Overall, the analysis provides a structured and critical overview of data-driven AQI forecasting methodologies and highlights their potential contribution to sustainable environmental monitoring and informed decision-making aligned with the United Nations Sustainable Development Goals (SDGs), including good health, sustainable communities, and climate action.</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

Deep Learning and Federated Learning in Air Quality Forecasting: Trends, Insights, Challenges, and Future Perspectives

  • 期刊论文

  • 发表日期：

    2026年4月4日

查看全文
  • Nairita Sarkar,
  • Pankaj Kumar Keserwani,
  • Rakesh Ranjan,
  • Mahesh Chandra Govil

摘要

Air is a vital natural resource that sustains life by providing oxygen and maintaining the balance of the Earth’s ecosystems. Clean air supports human health, climate regulation, agricultural productivity, urban planning, and overall well-being. However, increasing concentrations of harmful gases and particulate matter released through anthropogenic and natural activities have significantly degraded air quality in many regions worldwide. Accurate forecasting of air pollution levels has therefore become essential for effective environmental monitoring and public health protection. In recent years, Artificial Intelligence (AI)-driven approaches have gained prominence for modeling the complex and non-linear dynamics of air pollution. The article presents a comprehensive literature survey on air quality index (AQI) prediction using standalone deep learning (DL) models, hybrid DL architectures, and federated learning (FL) frameworks. The survey systematically reviews pollutant characteristics, meteorological factors influencing air quality, commonly used datasets, data preprocessing strategies, and performance evaluation metrics reported in recent studies. Further, this review article highlights key methodological challenges, comparative performance trends, and emerging research directions, with particular emphasis on privacy-preserving and decentralized learning paradigms enabled by FL. Overall, the analysis provides a structured and critical overview of data-driven AQI forecasting methodologies and highlights their potential contribution to sustainable environmental monitoring and informed decision-making aligned with the United Nations Sustainable Development Goals (SDGs), including good health, sustainable communities, and climate action.