A Study on ANN Performance Towards Three Significant Compounds of High Quality Agarwood Oil

Abstract

This study demonstrated the application and the performance of the artificial neural network (ANN) as classification tool for luxury oil which is agarwood essential oil. For the scope of this research, the compounds of agarwood essential oil were obtained from FRIM and BARCE (UMP). The 103 compounds data is pre-processed through a pre-processing technique known as principal component analysis (PCA) and Pearson’s correlation. It was found that three compounds were significant and they were high quality; β-agarofuran, α-agarofuran, and 10-epi-ϒ-eudesmol. The significant compounds were continued to be fed into ANN as input data meanwhile the output data categorized as low and high quality of the agarwood essential oil. The Scaled Conjugate Gradient (SCG) was employed as the default classifier algorithm during network training. Three layers of ANN architecture were used and 1 to 10 hidden neurons were varied in a hidden layer. The performance of the ANN was measured using the mean squared error (MSE), epochs and their execution time and the confusion matrix. The work was performed using Matlab R2017a. The finding shows that SCG-ANN successfully classified agarwood essential oil with the best performance at 3 hidden neurons. This research is significant for future work, especially on the classification of the agarwood essential oil field.