Estimation of asphaltene adsorption on MgO nanoparticles using ensemble learning

Abstract

The purpose of this study is to develop a reliable predictive model using machine learning algorithms that will predict asphaltene adsorption on MgO nanoparticles. Three machine learning (ML) algorithms namely gradient boosting machine (GBM), bagging (BG), and random forest (RF) have been applied for the development of models. The models have been constructed using 36 data points consisting of four features which include nanoparticle surface area, nanoparticle dosage, experimental temperature, and equilibrium concentration of asphaltene in the supernatant (Toluene). Correlation analysis is carried out using a heat map. It has been found that adsorption of asphaltene on MgO nanoparticle is considerably increased by increasing the surface area of nanoparticles while highly decreased by increasing the dosage of the nanoparticle. Temperature and equilibrium concentration showed a low effect on the adsorption phenomenon. All developed models are validated using a new experimental dataset that is not utilized during the model development process. For finding the accuracy of all models, different statistical parameters are evaluated and comprehensive graphical error analysis has been employed. The results depict that all ML algorithms perform well but GBM forecast highly accurate results as compared to other models. During the model development process, the performance of models is found in order: GBM ​> ​RF ​> ​BG while in the validation step the accuracy of models is determined in the following sequence: GBM ​> ​BG ​> ​RF. Furthermore, the leverage approach has also been applied for the detection of outliers. Fortunately, all experimental data is found reliable (in the applicability domain) indicating that developed models are reliable for predicting asphaltene adsorption on MgO nanoparticles.