Scholarlink

<p>This study applies advanced machine learning techniques to forecast the Air Quality Index (AQI) in Phoenix, Arizona, USA, utilizing a comprehensive dataset spanning 24 years. The methodology centers on deploying the Random Forest (RF) and MLP models with Hyperparameter Optimization and advanced feature engineering to enhance predictive performance. A key innovation is the systematic use of lagged and rolling statistical features to capture the dynamic nature of air quality fluctuations. Lag features, representing AQI values from the previous 1 to 7 days, and rolling statistics, including 3-day and 7-day rolling means and standard deviations, were instrumental in capturing short-term trends and temporal dependencies in AQI. These features reflect the influence of local emission patterns and meteorological conditions, with analysis highlighting the strong predictive power of short-term features like the 3-day rolling mean. Feature importance analysis and dimensionality reduction addressed multicollinearity, enhancing model efficiency and reliability. The optimized MLP model outperformed RF, achieving an MSE of 0.04 on unseen data. This study provides practical tools for real-time AQI forecasting, enhancing environmental learning, and supporting public health and urban planning. It highlights the importance of lagged and rolling features, establishing a robust framework and benchmark for future air quality predictive modeling.</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

Advancing air quality prediction with hyperparameter optimization and innovative feature analysis using deep learning models in Phoenix, Arizona, USA

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Vijendra Kumar,
  • Akshat Agrawal,
  • Naresh Kedam,
  • Saleh Alsulamy,
  • Aditya Singh

摘要

This study applies advanced machine learning techniques to forecast the Air Quality Index (AQI) in Phoenix, Arizona, USA, utilizing a comprehensive dataset spanning 24 years. The methodology centers on deploying the Random Forest (RF) and MLP models with Hyperparameter Optimization and advanced feature engineering to enhance predictive performance. A key innovation is the systematic use of lagged and rolling statistical features to capture the dynamic nature of air quality fluctuations. Lag features, representing AQI values from the previous 1 to 7 days, and rolling statistics, including 3-day and 7-day rolling means and standard deviations, were instrumental in capturing short-term trends and temporal dependencies in AQI. These features reflect the influence of local emission patterns and meteorological conditions, with analysis highlighting the strong predictive power of short-term features like the 3-day rolling mean. Feature importance analysis and dimensionality reduction addressed multicollinearity, enhancing model efficiency and reliability. The optimized MLP model outperformed RF, achieving an MSE of 0.04 on unseen data. This study provides practical tools for real-time AQI forecasting, enhancing environmental learning, and supporting public health and urban planning. It highlights the importance of lagged and rolling features, establishing a robust framework and benchmark for future air quality predictive modeling.