Feature-driven Hybrid Attention Learning for Accurate Water Quality Prediction
Published in Expert Systems with Applications (ESWA), 2025
This study presents a novel feature-driven hybrid attention model for accurate water quality prediction. It combines the Analytic Hierarchy Process (AHP) using linguistic terms with weakened hedges for robust feature selection, with a dual-attention mechanism integrating graph attention and Informer-based self-attention modules.
Applied to datasets from the Yellow River Basin, the model outperforms eight baselines with up to 68% improved accuracy and 91.6% reduction in error. This framework provides critical insights for sustainable water management under climate change scenarios.
Download paper from ESWA.
Recommended citation: X. Yao, Z. Xu, T. Ren, and X.-J. Zeng, “Feature-driven Hybrid Attention Learning for Accurate Water Quality Prediction,” Expert Systems with Applications, vol. 276, 127160, 2025.
Recommended citation: X. Yao, Z. Xu, T. Ren, and X.-J. Zeng, "Feature-driven Hybrid Attention Learning for Accurate Water Quality Prediction," Expert Systems with Applications, vol. 276, 127160, 2025.
Download Paper