Hybrid Swin Transformer–BiGRU Framework for Spatio-Temporal Prediction of Urban Air, Water, and Waste Dynamics in Smart Cities

Accurate and timely prediction of environmental dynamics—such as air pollution levels, water quality degradation, and urban waste accumulation—is critical for the sustainable planning and management of smart cities. However, conventional predictive frameworks often fall short in capturing the intricate spatial interdependencies and long-range temporal sequences inherent in heterogeneous sensor networks deployed across urban areas. To address these limitations, this study proposes a novel hybrid deep learning architecture that integrates a Swin Transformer for spatial feature learning with a Bidirectional Gated Recurrent Unit (BiGRU) for temporal sequence modelling. The system processes diverse sensor modalities, including PM2.5 concentration, CO2 levels, water pH and turbidity, and smart waste bin fill levels. These inputs are mapped into geospatial grids representing real-world urban zones, facilitating structured spatio-temporal analysis. Extensive experimentation on real-world smart city datasets demonstrates that the proposed Swin–BiGRU framework outperforms existing baselines, including CNN–LSTM, standalone GRU, and Transformer-based models. The model achieves consistently lower values for Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), while attaining higher values for the Pearson correlation coefficient (R) and the coefficient of determination (R²) in all predictive tasks. These results underscore the model’s enhanced ability to capture complex spatio-temporal patterns and its potential for reliable deployment in real-time urban environmental monitoring systems.