A systematic and critical review of the application of machine learning algorithms in modeling nanoparticle transport in porous media

Growing concerns about the potential risk for human exposure to nanoparticle transport and their consequences have drawn researchers' attention to studying the transport behavior in porous media under various conditions. Understanding of this behavior requires the development of reliable predictive models. Therefore, the main objective of this study is to systematically and critically investigate different machine learning-based models from different perspectives to predict the transport efficiency of a wide range of nanoparticles and identify important features. To achieve this goal, first, published articles for selected nanoparticles were collected. Then, machine learning algorithms including linear regression (LR), k-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), random forest (RF), gradient boosting regression (GBR), and artificial neural networks (ANN) were reviewed. In most of these algorithms, retention rate is considered as the response variable and particle, porous media, and flow characteristics are considered as predictor variables. The results of the reviewed studies showed that the random forest (27%), artificial neural network (23%) and decision tree (14%) algorithms are most useful in predicting the transport of nanoparticles in porous media. The importance analysis of the features indicates that five features including ionic strength, pore volume, zeta potential of particles, input concentration and diameter of nanoparticles exert the greatest influence. The predictive algorithms mentioned in this study can be used to control and manage the release of nanoparticles used in various industries into the environment and reduce their health risks.