<![CDATA[General Background: Air pollution has become a critical global issue affecting environmental sustainability and public health, creating a strong demand for accurate air quality prediction systems. Specific Background: Traditional statistical models and conventional machine learning techniques often struggle to capture the nonlinear and multivariate characteristics of environmental data, particularly when dealing with complex temporal dependencies. Knowledge Gap: Many existing forecasting approaches focus primarily on either short-term sequential learning or long-range temporal modeling, which limits their ability to represent both bidirectional temporal patterns and long-term dependencies in multivariate air quality datasets. Aims: This study proposes a hybrid deep learning framework integrating Transformer, Bidirectional Long Short-Term Memory (BiLSTM), and an Attention mechanism for accurate multivariate air quality prediction. Results: Experiments conducted on the UCI Air Quality dataset demonstrate that the proposed model achieves superior predictive performance with RMSE of 0.0799, MAE of 0.0589, and R² of 0.9621, outperforming baseline models such as standalone Transformer and BiLSTM architectures. Novelty: The proposed framework combines global temporal dependency modeling from Transformer encoders with bidirectional sequence learning from BiLSTM and adaptive temporal weighting through the attention mechanism. Implications: The framework provides a reliable computational approach for environmental monitoring systems, supporting intelligent air quality forecasting, early warning mechanisms, and data-driven environmental decision-making. Highlights Hybrid architecture captures both long-range temporal dependencies and bidirectional sequence relationships in environmental data. Multivariate forecasting shows strong predictive consistency across several pollutants and meteorological variables. Experimental evaluation reports very low prediction errors and strong statistical correlation with observed measurements. Keywords: Air Quality Prediction, Multivariate Time Series, Hybrid Deep Learning, Transformer BiLSTM Model, Environmental Monitoring]]>
Copyrights © 2026
