Application of KNN and Semi-Empirical Models for Prediction of Polycyclic Aromatic Hydrocarbons Solubility in Supercritical Carbon Dioxide

The solubility of polycyclic aromatic hydrocarbons in supercritical carbon dioxide is of great importance in a wide range of applications such as extraction from polluted soils and catalytic hydrogenation. In this study, for the first time a K-nearest neighbor (KNN) as a nonparametric and easy learning model was proposed to predict the solubility of 11 polycyclic aromatic hydrocarbons in supercritical carbon dioxide. The KNN model was constructed with optimum values of K for different compounds. To have a better understanding of the KNN model capability, predictions were compared with the outputs of several well-known semi-empirical models (Chrastil, Adachi and Lu, Li, Garlapati and Madras, Sparks and Bian). Results showed that the KNN model has an average absolute relative deviation (AARD) of about 12.47% for 11 polycyclic aromatic hydrocarbons modified density based correlations (Sparks and Bian) while the best semi-empirical equation (Sparks) result an AARD of about and 11.50%.