application of genetic algorithm based support vector machine model in second virial coefficient prediction of pure compounds

In this work, a genetic algorithm boosted least square support vector machine model by a set of linear equations instead of a quadratic program, which is improved version of support vector machine model, was used for estimation of 98 pure compounds second virial coefficient. compounds were classified to the different groups. finest parameters were obtained by genetic algorithm method for training data. the accuracy of the genetic algorithm boosted least square support vector machine was compared with four empirical equations that are well-known and are claimed can predict all compounds second virial coefficients (pitzer, tesonopolos, gasanov rk and long meng). results showed that in all classes of compounds, the genetic algorithm boosted least square support vector machine method was more accurate than these empirical correlations. the average relative deviation percentage of overall data set was 2.53 for the genetic algorithm boosted least square support vector machine model while the best average relative deviation percentage for empirical models (tesonopolos) was 15.38. when the molecules become more complex, the difference in accuracy becomes sharper for empirical models where the proposed genetic algorithm boosted least square support vector machine model have predicted good results for classes of compounds that empirical correlations usually fail to give good estimates.