THE PREDICTION OF KOVATS RETENTION INDICES OF ESSENTIAL OILS AT GAS CHROMATOGRAPHY USING GENETIC ALGORITHM-MULTIPLE LINEAR REGRESSION AND SUPPORT VECTOR REGRESSION

The Kovats retention indices of 340 essential oil compounds have been successfully predicted based on their molecular descriptor using the Multiple Linear Regression (MLR) and Support Vector Regression (SVR). The genetic algorithm (GA) was used to select the best molecular descriptors, resulting in the selection of the five best molecular descriptors to construct the Kovats retention index prediction model. As the results, MLR had R-2 training = 0.970, R-2 testing = 0.970, RMSE training = 56.55, and RMSE testing = 56.99. Meanwhile SVR model produced R-2 training = 0.981, R-2 testing = 0.973, RMSE training = 44.62 and RMSE testing = 53.60. The MLR model obtained the average difference of the predicted values as 3.8% for the training set and 3.4% for the testing set. Meanwhile, SVR yielded a 2.4% difference for the training set and 3.4% for the testing set. Compared to MLR, the SVR model gave higher R-2, lower RMSE, and a lower average difference between the predicted and observed values. In conclusion, our results indicate that the SVR is a more accurate predictor of the Kovats retention index than the MLR.