Air quality prediction with a hybrid CNN-BiLSTM-attention model using multivariate environmental data

Accurate air quality forecasting is important in mitigating the adverse health effects of air pollution; it provides a guideline necessary in driving public health policy. The new hybrid model developed here is the CNN-BiLSTM-Attention-Dense algorithm, and it has been applied for a 30-day ahead forecast using the historical data of the period 2014-2024. The dataset covers several air pollutants such as PM2.5, NO, PM10, NO2, NH3, NOx, SO2, CO, O3, Xylene, Toluene, and all other meteorological variables that include temperature, humidity, wind speed, wind direction, and solar radiation. This work proposes a model with Convolutional Neural Networks that will capture the features of a sample, a network of Bidirectional Long Short-Term Memory to get hold of the temporal dependencies within these activities, and multi-head attention that shall dynamically give emphasis to critical aspects of interest within the data. The results of extensive experiments prove that the proposed CNN-BiLSTM-Attention-Dense model gives much better performance compared to traditional models such as CNN, LSTM, CNN-LSTM, and CNN-LAD in terms of MSE, RMSE, and MAPE. For 30-day AQI prediction, the proposed model has secured an MSE of 0.000982, RMSE of 0.03134, and MAPE of 23.5487, which ensures high accuracy and robustness of the proposed approach. Symposium This research work has made it possible for real-time and accurate forecast of air quality that can be greatly beneficial for environmental officials and policy decision-makers to make wise decisions towards protection of public health, as well as for formulation of enhanced air quality management plans.