The k-Nearest Neighbor modelling by varying Mahalanobis and Correlation in distance metric for agarwood oil quality classification

Abstract

Agarwood oil is well known for its unique scent and has many usages; as an incense, as ingredient in perfume, is burnt during religious ceremonies and is used in traditional medical preparation. Therefore, agarwood oil has high demand and is traded at different price based on its quality. Basically, the oil quality is classified by using physical properties (odor and color) and this technique has several problems: not consistent in term of accuracy. Thus, this study presented a new technique to classify the quality of agarwood oil based on chemical properties. The work focused on the k-Nearest Neighbor (k-NN) modelling by varying Mahalanobis and Correlation in distance metric for agarwood oil quality classification. It involved of 96 samples of agarwood oil, data pre-processing (data randomization, data normalization, and data division to testing and training datasets) and the development of k-NN model. The training dataset is used to train the k-NN model, and the testing dataset is used to test the developed model. During the model development, Mahalanobis and Correlation are varied in k-NN distance metric. The k-NN values are ranging from 1 to 10. Several performance criteria including resubstitution error (closs), cross-validation error (kloss) and accuracy were applied to measure the performance of the built k-NN model. All the analytical work was performed via MATLAB software version R2020a. The result showed that the accuracy of Mahalanobis distance metric has a better performance compared to Correlation from k=1 to k=5 with the value of 100.00%. This finding is important as it proved the capabilities of k-NN modelling in classifying the agarwood oil quality. Not limited to that, it also contributed to the agarwood oil research area as well as its industry.