Scholarlink

<p>Atmospheric pollution is a complex and multi-faceted issue, especially when it comes to obtaining accurate, real-time data for reliable forecasting. This study evaluates the performance of six machine learning (ML) methods, Multiple Linear Regression (MLR), Random Forest (RF), Support Vector Machine (SVM), Feedforward Neural Networks (FNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU), for predicting hourly ozone levels in Casablanca, Morocco. The dataset comprises 1,320 hourly observations collected in Casablanca, encompassing 12 environmental and meteorological variables, including ozone (O<sub>3</sub>), carbon monoxide (CO), carbon dioxide (CO<sub>2</sub>), nitrogen oxides (NO<sub>x</sub>), particulate matter (PM<sub>2.5</sub> and PM<sub>10</sub>), temperature, wind speed, wind direction, and humidity. Among the models tested, Random Forest (RF) achieved the highest predictive accuracy, with an RMSE of 9.18 and a correlation coefficient of 91.77%. Its strength lies in handling medium-sized datasets with minimal preprocessing. While RF successfully captured overall ozone trends, it struggled with sharp peaks and rapid changes. To improve this, the study introduced a second phase using hybrid models that combine RF with secondary learners like FNN, SVM, LSTM, and GRU. The RF-FNN hybrid produced the most accurate and well-timed forecasts, with an RMSE of 2.50 and a correlation coefficient of 99.29%. These results suggest that combining tree-based models with neural networks can significantly enhance short-term air quality forecasting.</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

Improving the forecasting of air pollutant gases applying hybrid machine learning methods: application to extremes

  • 期刊论文

  • 发表日期：

    2025年9月30日

查看全文
  • Anas Adnane,
  • Amine Ajdour,
  • Radouane Leghrib

摘要

Atmospheric pollution is a complex and multi-faceted issue, especially when it comes to obtaining accurate, real-time data for reliable forecasting. This study evaluates the performance of six machine learning (ML) methods, Multiple Linear Regression (MLR), Random Forest (RF), Support Vector Machine (SVM), Feedforward Neural Networks (FNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU), for predicting hourly ozone levels in Casablanca, Morocco. The dataset comprises 1,320 hourly observations collected in Casablanca, encompassing 12 environmental and meteorological variables, including ozone (O3), carbon monoxide (CO), carbon dioxide (CO2), nitrogen oxides (NOx), particulate matter (PM2.5 and PM10), temperature, wind speed, wind direction, and humidity. Among the models tested, Random Forest (RF) achieved the highest predictive accuracy, with an RMSE of 9.18 and a correlation coefficient of 91.77%. Its strength lies in handling medium-sized datasets with minimal preprocessing. While RF successfully captured overall ozone trends, it struggled with sharp peaks and rapid changes. To improve this, the study introduced a second phase using hybrid models that combine RF with secondary learners like FNN, SVM, LSTM, and GRU. The RF-FNN hybrid produced the most accurate and well-timed forecasts, with an RMSE of 2.50 and a correlation coefficient of 99.29%. These results suggest that combining tree-based models with neural networks can significantly enhance short-term air quality forecasting.