A two-stage residual modeling framework for data-driven prediction of uranium adsorption performance in aqueous environments.

A two-stage residual modeling framework for uranium adsorption performance in aqueous environments improves prediction accuracy, with MLP showing a 33.7% increase in R² and 44.4% lower RMSE.

• Why it matters: Uranium contamination threatens radioactive wastewater treatment, and existing theoretical/experimental modeling is insufficient for efficient, data-driven adsorbent design under laboratory conditions.
• What they did: Intrinsic material characteristics and uranium adsorption parameters were identified, then nine machine learning algorithms were optimized via Bayesian optimization and cross-validation. A residual optimization stage used a well-trained Random Forest, and SHAP plus PDP interpreted the residual model.
• The result: The MLP’s statistically significant gains indicate adaptive correction of prediction biases across algorithms, enabling rapid assessment of uranium adsorption behavior under laboratory conditions.

AI summary of the abstract · may contain errors · not medical advice · read the paper.