Evolving a least square support vector machine using real coded shuffled complex evolution for property estimation of aqueous ionic liquids

Abstract

In this study, we demonstrate how least square support vector machine (LSSVM) evolution with the shuffled complex evolution (SCE) ameliorates the predictability and reliability of the support vector machine as an estimation tool for thermodynamic of ionic liquids solutions. This strategy is applied to forecast the osmotic-coefficient of the 26 different ionic liquids by utilizing the 1409 available archival literature data points. Our methodology is the development of a hybrid SCE-LSSVM algorithm. Shuffled complex evolution is used to decide the hyper parameters of support vector machines so that all the initial weights can be searched and obtained intelligently. The evolution operators and parameters are carefully designed and set to avoid premature convergence and permutation problems. The results demonstrate that carefully designed SCE-LSSVM outperforms the structural risk minimization of support vector machines, predicting the properties of aqueous solutions in a way, even better than the available models.