Prediction of novel ionic liquids’ surface tension via Bagging KNN predictive model: Modeling and simulation

In recent decades, development of computational approaches to theoretically estimate the surface tension of various types of ionic liquids (ILs) has allowed the scientists to appropriately arrange new industrial-based processes with higher cost-effectiveness compared to experimental methods. In this paper three artificial intelligence (AI)-based predictive models have been developed to measure the surface tension two prevalent ILs including 1,3-dimethylimidazolium methylsulfate and 1-ethyl-3-methylimidazolium ethylsulfate. To create prediction models in this study, we employed Lasso, K Nearest Neighbor (KNN), and Linear Regression (LR) as basis models of a bagging regression ensemble. However, the dataset used has 45 input features and one output feature (surface tension). To find most important and useful input features, we also require a feature selection mechanism. We used ant colony optimization as a search strategy to discover the best feature combination for this purpose. Finally, the models were tested using various standard metrics. Bagging LR, Bagging Lasso, and Bagging KNN illustrate R-squared of 0.934, 0.915, and 0.977. Bagging KNN has been selected as the main model in this investigation because it had less mistakes and a higher R-squared.