Kovats retention index is a retention parameter used to identify analytes in gas
chromatography. The value of the Kovats retention index in compounds that have
similarities tends to have a similar pattern. It may show that there is a strong
relationship between the chemical structure and the value of the Kovats retention
index. The Fuzzy Support Vector Regression (FSVR) method is one of the algorithms
that can be used to calculate the retention index of Kovats value out of 51 flavor and
fragrance compounds. The program is built using the Python programming language.
The calculation of the index value begins by inputting the molecular descriptor data
into the sigmoid fuzzy membership function. The 51 compounds were then divided into
2 parts, namely 80% of the data into the training set and 20% of the data into the
testing set. From the training data, the hyperparameter tuning process is then carried
out to get the best parameters to build the model. The kernels used are linear kernel
and Radial Basis Function (RBF) kernel. The linear kernel uses C and epsilon
parameters, each of which has 7 values, resulting in 49 combinations of parameters.
The RBF kernel uses parameters C, gamma, and epsilon, each of which has 7 values,
resulting in 343 combinations of parameters. Out of all types of parameter
combinations, a linear kernel with parameter values of C = 1000 and Epsilon = 10
gives the best results. These parameters provide an optimal value with an R-Square of
0.977 and an RMSE of 37,215. The FSVR model can predict the value of the Kovats
retention index on the testing data with an average difference of 2.8%. These results
prove that the linear kernel is better than the RBF kernel in predicting the value of the
retention index of Kovats and the FSVR method can be used as a method to predict
retention index of Kovats value on flavor and fragrance compounds.