A Multimodal AI Framework for Real-Time Air Quality Prediction and Pollution Source Identification in Smart Cities

Abstract

Air pollution is a major problem for smart cities. Typically, traditional methods of monitoring air pollution lack the ability to properly predict the complex temporal features of the data as well as account for the meteorological factors and determine pollution sources. This research proposes a Multimodal AI Framework based on a combination of Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM) and XGBoost to predict air quality and identify sources of pollution. The pollutant data is combined with meteorological variables to extract spatial patterns, temporal dependencies and non-linear relationships. This hybrid approach aims to provide better prediction accuracy compared to traditional machine learning and deep learning approaches by combining the benefits of both. In an example, the model achieved an MAE of 7.8 µg/m³, RMSE of 12.4 µg/m³, R² of 0.94, MAPE of 12.6% and an inference time of 18ms. These results will need to be verified experimentally.