Scholarlink

<p>This study evaluates the effectiveness of the Random Forest (RF) model combined with four feature selection (FS) algorithms including the Sequential Forward Selection (SFS), Sequential Backward Selection (SBS), Recursive Feature Elimination (RFE), and Boruta for predicting PM2.5 and PM10 air quality indices (AQI) in urban areas of central Iran. Over 44% of days during the study period experienced unhealthy air conditions. The models were trained on 70% of the data using different feature selection methods and validated on the remaining 30%. Permutation Feature Importance (PFI) method was finally utilized to determine the importance of factors affecting daily changes in AQIPM2.5 and AQIPM10. RF-SFS (R<sup>2</sup> = 0.56) and RF-SBS (R<sup>2</sup> = 0.58) showed the highest performance for predicting AQIPM2.5 and AQIPM10, respectively. The PFI results revealed that maximum wind speed (21.4%), horizontal visibility (19.1%), sunny hours (13.8%), and wind speed (10%) were the main factors influencing AQIPM2.5 variations. However, horizontal visibility (19.7%), wind direction (10.7%), minimum air temperature (10.5%), and mean daily wind speed (10.3%) were the key factors controlling AQIPM10. These results can serve as a useful tool for enhancing air quality management strategies in the regions.</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

A feature-centric machine learning framework for forecasting urban air pollution in arid regions

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Zohre Ebrahimi-Khusfi,
  • Narjes Okati

摘要

This study evaluates the effectiveness of the Random Forest (RF) model combined with four feature selection (FS) algorithms including the Sequential Forward Selection (SFS), Sequential Backward Selection (SBS), Recursive Feature Elimination (RFE), and Boruta for predicting PM2.5 and PM10 air quality indices (AQI) in urban areas of central Iran. Over 44% of days during the study period experienced unhealthy air conditions. The models were trained on 70% of the data using different feature selection methods and validated on the remaining 30%. Permutation Feature Importance (PFI) method was finally utilized to determine the importance of factors affecting daily changes in AQIPM2.5 and AQIPM10. RF-SFS (R2 = 0.56) and RF-SBS (R2 = 0.58) showed the highest performance for predicting AQIPM2.5 and AQIPM10, respectively. The PFI results revealed that maximum wind speed (21.4%), horizontal visibility (19.1%), sunny hours (13.8%), and wind speed (10%) were the main factors influencing AQIPM2.5 variations. However, horizontal visibility (19.7%), wind direction (10.7%), minimum air temperature (10.5%), and mean daily wind speed (10.3%) were the key factors controlling AQIPM10. These results can serve as a useful tool for enhancing air quality management strategies in the regions.