Dominant operating effects and localized material responses in dye removal by MOF based composites: Insights from interpretable machine learning.

Reliable comparison of MOF based composites for dye contaminated water treatment remains challenging because reported removal capacities do not reveal whether performance variations arise from broadly influential operating effects or material responses confined to specific treatment contexts. Here, we developed an interpretable machine learning framework using heterogeneous literature derived experimental records to distinguish broadly influential operating effects from localized material responses. The compiled observations were organized within a unified descriptor space and refined before model development, while representative dye-framework interactions were subsequently examined at the molecular level to support the interpretation of material-dependent responses. Among the evaluated regression algorithms, tree based ensemble models provided the most reliable representation of capacity variations, with XGBoost and XGBoost-CMAES showing the best overall performance. SHAP and interaction analyses revealed that initial concentration, contact time, and adsorbent dosage showed the broadest model derived contribution across the dataset. In contrast, specific surface area and pore size contributed more selectively, with their effects becoming pronounced within particular response regions and in combination with dye related characteristics. Temperature, salt concentration, morphology factor, and zeta potential showed comparatively limited contributions within the sampled domain. Molecular analysis of representative dyes and aromatic MOF related fragments indicated that π-π stacking stabilized favorable binding configurations, providing molecular-level support for the observed material-dependent responses. By separating dataset wide operating effects from condition dependent material responses, this work moves beyond overall feature ranking and provides a clearer basis for evaluating MOF based composites under heterogeneous dye removal conditions.