Table 10 Proposed model comparisons with other state-of-the-art air quality forecasting models.
From: Advanced air quality prediction using multimodal data and dynamic modeling techniques
Reference | Method | Accuracy | RMSE | MAE | R2 | Key Features |
|---|---|---|---|---|---|---|
Proposed hybrid model | CNN + BiLSTM + Attention + GNN + Neural-ODE | 98.8% | 6.21 | 3.89 | 0.988 | Combining CNNs, BiLSTM, attention mechanisms, GNNs, and Neural-ODEs to handle spatial and temporal features |
Ahmed et al.1 | Deep Learning for AQI Forecasting with Satellite-derived Data | 90.32% | 7.10 | 4.30 | 0.96 | Uses hydro-climatological data along with deep learning for AQI forecasting |
Rabie et al.2 | CNN-BiLSTM Hybrid for Air Quality Prediction | 92.73% | 8.05 | 4.20 | 0.94 | Utilises CNN and BiLSTM hybrid for spatially resolved AQI prediction in megacities |
Kumar & Kumar3 | Multi-view Stacked CNN-BiLSTM for PM2.5 Prediction | 91.62% | 7.80 | 4.50 | 0.95 | Stacked CNN-BiLSTM approach to predict PM2.5 concentrations in urban environments |
Putri & Caraka (2024) | Fine-tuning CNN-LSTM for PM2.5 Nowcasting | 92.72% | 7.65 | 4.35 | 0.95 | Focuses on fine-tuning CNN-LSTM models for short-term PM2.5 predictions |
Barthwal & Goel4 | DCNN + LSTM for AQI Time-series Classification | 90.16% | 8.10 | 4.55 | 0.93 | Integration of DCNN and LSTM for AQI time-series classification |
Wu et al.5 | DVMD Informer-CNN-LSTM Optimized with Dung Beetle Algorithm | 91.82% | 7.00 | 4.00 | 0.96 | Optimised with the Dung Beetle Algorithm for enhanced AQI forecasting performance |
Prado-Rujas & García-Dopico6 | Spatio-Temporal Air Quality Forecasting Framework | 93.22% | 8.20 | 4.60 | 0.92 | Multi-variable, sensor-agnostic framework for spatio-temporal air quality forecasting |
Liu et al.7 | Air Quality Class Prediction using ML | 91.29% | 8.40 | 4.70 | 0.91 | Uses machine learning for air quality classification based on monitoring and secondary data |
Wang et al.8 | Spatio-Temporal Model Integrating Graph Convolution | 93.08% | 6.80 | 4.10 | 0.97 | Combines graph convolution with multi-head attention mechanisms for improved air quality forecasting |
