Performance Evaluation of Distance Function in KNN and WKNN for Classification of Soil Organic Matter

Abstract

Essential oils plantation requires sufficient nutrients and organic matter to produce high quality oil. This study introduces a new technique in identifying soil organic matter content by utilizing soil images and nearby distance-based classification algorithms such as KNN and Weighted KNN. However, the problem with distance-based algorithms is the use of distance calculation techniques, which can produce significant differences in calculations. Therefore, in this study a test was conducted on the use of different distance formulation, namely Manhattan, Euclidean and Minkowsky. The features used as input to the recognition process were extracted using the Gray Level Co-occurence Matrix (GLCM) method. The test results on 847 training data and 363 test data showed that the best distance function in K-Nearest Neighbor (KNN) and Weighted-KNN (WKNN) was Minkowsky with a rating value (r) of 5. This shows that the selection of the right distance function will increase the accuracy of the KNN and WKNN