Transformer-Based Encoder-Decoder Model For Enhanced Air Quality Prediction

Air quality prediction remains a critical challenge due to the complex spatiotemporal dependencies inherent in pollutant data. We propose a transformer-based encoder-decoder model to address this challenge, focusing on accurate and robust air quality index (AQI) forecasting. The proposed method processes historical pollutant measurements, including PM2.5, PM10, and NO2, through a multi-head self-attention mechanism to capture long-range dependencies and nonlinear interactions. The model employs a sliding window approach to generate sequential input-output pairs, which are then normalized and fed into stacked transformer encoders for feature extraction. A global average pooling layer condenses the temporal information into a fixed-length representation, enabling precise AQI prediction through a dense output layer. The architecture incorporates residual connections and layer normalization to stabilize training, while dropout regularization mitigates overfitting. Experiments on the Delhi air quality dataset demonstrate the model’s effectiveness, achieving competitive performance in terms of mean squared error and mean absolute error. Furthermore, the transformer’s ability to model intricate temporal patterns without recurrent structures offers computational advantages over traditional sequence models. The results highlight the potential of attention-based architectures for environmental monitoring tasks, particularly in scenarios where interpretability and scalability are paramount. This work contributes to the growing body of research on deep learning for air quality prediction, providing a framework that balances accuracy, efficiency, and generalizability