Comparison of ANN Performance Towards Agarwood Oil Compounds Pre-processing Based on Principal Component Analysis (PCA) and Stepwise Regression Selection Method

This paper presents the performance of Artificial Neural Network (ANN) application towards the agarwood oil quality classification. The works involved the uses of agarwood oil compounds based on two different feature selection techniques. The compounds were are selected based on using Principal Component Analysis (PCA) and Stepwise Regression. The compounds identified by PCA (three compounds) were β-agarofuran, α-agarofuran, and 10-epi-ϒ-eudesmol while the compounds identified by stepwise regression (four compounds) were β-agarofuran, ϒ-Eudesmol, Longifolol, and Eudesmol. These compounds were fed into ANN separately as input features and the output was the quality of the oil either high and low. The Resilient Backpropagation as classifier algorithm was used and 1 to 10 hidden neuron in the hidden layer were varied. The performance of ANN using three and four compounds was measured and compared using confusion matrix, mean square error (mse) value and number of epoch. The work was done using software application, Matlab R2017a by using ‘patternet’ network. The finding showed that the ANN using four compounds of agarwood oil as input feature obtained greater performance with good accuracy, lower mse value and lower number of epoch in one hidden neuron.