Adsorptive removal rate prediction of imprinted polymers for Cr(VI) by multivariate feature-engineered machine learning models and experimental validation

Adsorption prediction models of heavy metal IIPs by machine learning algorithms were constructed. • Multivariate feature engineering was applied according to structure-performance relationship. • The reliability of the prediction model was confirmed by experimental evaluation of prepared IIPs. • This study provided a data-driven model and profound mechanistic insights for rational design of IIPs. Ion-imprinted polymers (IIPs) are recognized as excellent adsorbents for heavy metal detection and removal to enable environmental safety. However, traditional trial-and-error development method for high performance IIPs is time-consuming, labor-intensive and costly. In this work, data-driven adsorptive removal rate prediction models were constructed which offers a promising alternative by facilitating rational design of IIPs structure and precise regulation of their adsorption performance. Specifically, multivariate feature engineering was performed to applied machine learning algorithms (Gaussian process regression, generalized regression neural network, least squares boosting, extreme gradient boosting, convolutional neural network, and multilayer perceptron regression) according to structure-adsorption performance relationships. Models were optimized via sparrow search algorithm, and their prediction performances were compared. Among them, extreme gradient boosting exhibited superior prediction accuracy and generalization, achieving correlation coefficient R 2 of 0.93 on training and testing sets, with RMSE and MAE low to 7.03 and 5.64, respectively. SHAP analysis revealed pH as the dominant influencing factor. Guided by this prediction model, novel Cr(VI) IIPs were designed and used for Cr(VI) adsorption from industrial wastewater with a removal efficiency of 72.75%-83.61%. The strong correlation (R 2 =0.79) between predicted and experimentally validated removal efficiency confirms the model's reliability and practical utility. This study not only provides novel high-performance IIPs for Cr(VI) adsorptive removal from environment, but also offers a reliable data-driven rational design strategy for functional materials development that can be used to ensure environmental safety. Download: Download high-res image (213KB) Download: Download full-size image