Abstract
Given the complex relationship between the structural features of carbon nanotube (CNT) membranes and their water permeability, predicting the performance of CNT membranes poses a significant challenge. The Bayesian optimization-based Extreme Gradient Boosting (Bayes-XGBoost) algorithm demonstrates considerable potential in capturing the intricate influences of various feature parameters on water permeability. An experimental dataset comprising 572 sets of data derived from molecular dynamics simulations serves as the characteristic dataset for machine learning, utilizing the Bayes-XGBoost algorithm to elucidate the connection between the structural features of CNT membranes and their filtration performance. The results indicate that, in predicting the permeability of CNT membranes, the Bayes-XGBoost algorithm achieves an impressive prediction accuracy of 97.82%, exhibiting faster convergence speed and higher predictive precision compared to traditional machine learning algorithms. Additionally, the optimal combination of CNT membrane feature parameters was identified through a genetic algorithm, providing robust support for the design and fabrication of high-performance CNT membranes. This highlights the significant potential of the Bayes-XGBoost in the field of material design.