Enhancing NARX Neural Network for Agarwood Oil Grading: A Study on Resilient Backpropagation Training Method

Agarwood oil, renowned both locally and internationally, is valued for its applications in perfumes, incense, and traditional medicine. However, it remains difficult to grade accurately due to a lack of precise methods. In the present study, data were obtained from the Forest Research Institute Malaysia (FRIM), Selangor Malaysia and BioAromatic Research Centre of Excellence (BARCE), Universiti Malaysia Pahang Al-Sultan Abdullah (UMPSA). This study, using data from a previous researcher, focuses on the resilient backpropagation (RBP) training function within the Nonlinear AutoRegressive eXogenous (NARX) model to evaluate its effectiveness for grading agarwood oil quality. An iterative approach is applied starting with simpler architectures and increasing complexity by adjusting the number of hidden neurons in the NARX model. The models are evaluated using performance metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), R-squared ($\mathbf{R}^{2}$), Epochs, and Accuracy to select the architecture that achieves the best performance. All analyses are conducted in MATLAB software version R2020a. The outcomes of this study will be highly beneficial for future studies in the field of agarwood oil, particularly in improving oil quality classification.