<![CDATA[Purpose – This study aims to develop a hybrid intelligent framework for prioritizing air quality monitoring stations by integrating multi-criteria decision-making (MCDM) methods with spatial graph-based deep learning to address limitations in interpretability and spatial dependency modeling. Design/methods/approach – The proposed framework combines Pythagorean Fuzzy Best–Worst Method (PF-BWM) to determine uncertainty-aware criteria weights and MARCOS to generate initial prioritization scores. A spatial graph is constructed by linking monitoring stations within a 50 km radius, and Graph Attention Networks (GAT) are applied to capture spatial dependencies. Final prioritization is obtained through a validation-based weighted fusion of MARCOS rankings and GAT predictions. The model is evaluated using air quality data from 247 monitoring stations across five countries obtained via the OpenAQ API (July–December 2025). Findings – The proposed framework achieves an accuracy of 93.8% and an F1-score of 93.1%, outperforming the standalone GAT model (92.6%). Statistical testing indicates that the improvement is significant (p = 0.003), demonstrating the effectiveness of combining MCDM and graph-based learning. Research implications/limitations – The framework enhances both interpretability and predictive performance in prioritizing monitoring stations. However, the study is limited by the selected spatial radius, dataset scope, and dependence on MARCOS-derived labels as ground truth. Future research should explore broader datasets and alternative graph construction strategies. Originality/value – This study presents a novel integration of uncertainty-aware MCDM (PF-BWM–MARCOS) with Graph Attention Networks, offering a more interpretable and spatially aware approach for air quality monitoring station prioritization.]]>
Copyrights © 2026
