On the evaluation of density of ionic liquids: towards a comparative study

•A new group contribution model was developed for predicting density of ionic liquids.•The model inputs are temperature, pressure, and 47 substructures.•A data bank containing 918 data points for 747 different ILs was used for the model.•Results indicate satisfactory predictions of suggested model than other existing models.•An outlier analysis was utilized to detect suspected data points. Superior physicochemical properties of ionic liquids (ILs) including dissolution potential for a large number of compounds, recyclability, suitable thermal stability, tuneability characteristics and trivial volatility make them attention-grabbing in electrochemistry and chemical industries. Owing to this fact, the accurate knowledge of ILs properties is demanded for the thermodynamic calculations involved in such processes. Amongst such properties, density is crucially significant in separation processes including CO2 absorption, extractive distillation and liquid–liquid extraction; thereby, creating and/or seeking a robust technique for density prediction is of great importance. In the present study, a new and combined version of least-square support vector machine as a powerful machine learning theory, and group contribution technique (GC-LSSVM) was extended for estimating the ILs density. It is worthwhile mentioning that genetic algorithm (GA) is applied to find the best values of kernel and regularization coefficients involved in GC-LSSVM. A widespread database was collected from the reliable open sources including 918 data points relevant to the 747 classes of ILs in relation with 47 substructures, pressure and temperature. The data was randomly separated into two subsets of test and train using a computer program. After the model was developed, graphical techniques and parametric statistics were executed to show the supremacy of the suggested GC-LSSVM in this study. The model findings were also compared to the available empirical and theoretical models in literature. Hence, the developed tool in this study gives the best match with target data and the least deviations from the actual ones with mean square error (MSE) of 0.0004 and coefficients of determination (R2) of 0.9925. The residual error analysis and outliers detection demonstrated the highest accuracy of GC-LSSVM model, and the validity of the employed database for density modeling, respectively. It can be concluded that the recommended tool in this study is a new combinatorial model which is employed for t