Abstract:Metal-organicframeworks (MOFs) have gained significantattention in the field of pollutant removal due to their rich porestructures and large specific surface areas. As the number of MOFstructures continues to increase, machine learning methods have becomea powerful tool for prediction of adsorptive activities of MOFs forpollutants. In this study, 16 models were constructed using publishedadsorption data, which included 28 MOFs and 30 pollutants, resultingin a dataset of 836 data points. The XGBoost model was determinedto be the most effective model, achieving an average R (2) of 0.953 during the 5-fold cross-validation. The model'sperformance was influenced by a combination of MOF features, pollutantfeatures, and adsorption conditions. Key parameters for the XGBoostmodel's performance included the pollutant concentration, pH,solid-liquid ratio, and temperature. Different types of MOFs,including Zr-MOFs, Cr-MOFs, Al-MOFs, and Fe-MOFs, were observed todisplay distinct adsorption mechanisms through the machine learningmodel. These mechanisms included electrostatic interactions, & pi;-& pi;interactions, hydrogen bonding, and van der Waals force. The model'spredictions regarding the optimal MOFs and adsorption conditions forthe 30 pollutants were partially validated through experimental data,demonstrating the feasibility of the model's predictions. Thisstudy provides technical and theoretical support for the predictionand selection of optimal MOFs for pollutant removal in the aqueousphase.

Prediction of Adsorptive Activities of MOFs for Pollutants in Aqueous Phase Based on Machine Learning