Residual Transformer Network Based Air Quality Forecasting with Hybrid Feature Extraction

Air quality prediction is one of the emerging research problems due to the increasing impact of atmospheric pollution on public health, urban sustainability and environmental governance. The traditional statistical and machine learning approaches has demonstrated the limited performance in handling the high dimensionality, nonlinear dynamics and temporal dependencies inherent in air quality data. The deep learning-based methods have demonstrated the improved forecasting accuracy with the challenges like redundant feature consideration, suboptimal solution and high computational cost. To address the limitations, this research proposed the optimized deep learning-based air quality prediction model with optimal attribute selection and deep learning-based air quality prediction. The publicly available air quality datasets are utilized by the proposed model and is employed with Z-score normalization for standardizing the data, optimal feature selection using Logistic Polar Light Optimization (LPL) algorithm, and Axial Tree Gated Recurrent Network (ATGNet) is employed for the air quality forecasting to enhance prediction performance. The proposed model demonstrated higher accuracy with improved generalization by reducing feature redundancy for reliable air quality forecasting.