Gru-lstm with attention-based forecasting for enhanced air quality

Air pollution, which is responsible for numerous chronic diseases and premature deaths, has become a growing global concern. Poor air quality not only harms human health and agriculture but also triggers severe political, social, and economic consequences. To analyze the state of the art, accurate environmental air pollution forecasting is thus essential to enable timely interventions, ensure public safety, and support informed policy decisions. This research proposes a robust and adaptable deep learning model Attention_GRU+LSTM for predicting ambient PM2.5 concentrations. The hybrid model integrates gated recurrent units (GRU) and long short-term memory (LSTM) networks with attention mechanisms to effectively capture both temporal and spatial dependencies in air quality data. Experimental validation was conducted using a publicly available Beijing air pollution dataset. The model was evaluated across short- and medium-term forecasting horizons (2-day, 5-day, and 10-day) using multiple performance metrics including Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE), and Coefficient of Determination (R2). For 2-day predictions, the proposed model achieved an RMSE of 10.735, MAE of 6.401, and R2 of 0.615, outperforming state-of-the-art models such as LSTM, GRU, and a hybrid deep learning model, whose RMSE values ranged from 12.568 to 14.107, MAE from 8.305 to 11.003, and R2 from 0.335 to 0.472. The model continued to show robust performance for longer-term forecasts, with RMSE of 24.779 (5-day) and 27.649 (10-day), maintaining its superiority over baseline and attention-based architectures. Furthermore, comparative results with models reported across previous studies confirm the consistent superiority and robustness of the proposed Attention_GRU+LSTM model.