| Abstrakt: |
This study examines the concentrations of particulate matter ≤ 2.5 μm (PM2.5) and sulfur dioxide (SO2) in Kerman city, Iran using three machine learning algorithms: random forest (RF), extreme gradient boosting (XGB), and monotonic multi-layer perceptron neural network (MMLP). Initially, a database of 22 influencing factors was gathered, from which 13 were selected based on the variance inflation factor for modeling. The target and selected variables were split into training (75%) and validation (25%) sets. Model performance was evaluated using the determination coefficient (R2), root mean square error (RMSE), mean absolute error (MAE), and nash sutcliffe efficiency coefficient (NSEC). The training and validation accuracy results for estimating levels of PM2.5 and SO2 with the RF model outperformed the XGB and MMLP models. Based on the reduction in RMSE (RMSER) following permutation feature importance (PFI), it was revealed that nitrogen dioxide (RMSER = 10), horizontal visibility (RMSER = 9.8), ozone (RMSER = 9.3), and air pressure (RMSER = 9.1) are the key factors influencing PM2.5 level variations in Kerman city. Additionally, it was noted that SO2 variations had a significant impact alongside three pollutants—nitrogen dioxide, ozone, PM2.5—and air pressure alterations in the city. The RMSE reductions after PFI for the aforementioned factors were 22.4, 19.9, 17.7, and 17.5, respectively, in relation to parameters affecting SO2. These results highlight the prominent role of these factors in changing SO2 concentrations in Kerman. Based on these findings, enhancing the planning and management of pollution sources can improve effectiveness in this urban area. |
Full text from SpringerLink