The kovats retention index obtained from 340 essential oil compounds has been sucessfully calculated using molecular descriptor data. The descriptor have been optimized using genetic algorithms and 5 most optimal descriptors are obtained as the result, namely ATSc1, VCH-7 SP-1, Kier1 and MlogP. This molecular descriptor will be used to predict the kovats retention index in essential oil compounds using the Python programming language and the Feed Forward Neural Network (FFNN) method using Particle Swarm Optimization (PSO) training. A total of 340 datasets will be divided into two combinations, namely training set and testing set with a ratio of 70:30 and 80:20. The test is carried out by finding the best FFNN model obtained by PSO using the ReLU activation function. The network model will differentiate each number of hidden layer neurons from 1 to 10, then each model test was carried out with 1000 iterations and 10 trials. From the test results, the best FFNN model that gives the most optimal results is the model with 3 hidden layer neurons that uses 80% training set and 20% testing set. The model gives the best prediction results with RMSE = 92.87, MAPE = 6.08% and R 2= 0,937. The average accuracy between the actual value and the predicted results of the kovats retention index on the test data is 93.9%. From the results obtained, it is known that the Feed Forward Neural Network (FFNN) method using Particle Swarm Optimization (PSO) training can be used as a method to predict the kovats retention index.
Keywords: kovats retention index, feed forward neural network, particle swarm optimization, molecular descriptors