Comparison of Distance Measurement Methods on K-Nearest Neighbor Algorithm For Classification

摘要

K-Nearest Neighbor is a non-parametric classification algorithm that does not use training data and initial assumptions or models in the calculation process. The quality of the k-Nearest Neighbor classification results is very dependent on distance between object and value of k specified, so the selection for distance measurement method determines the results of classification. This study compares several distance measurement method, including Euclidean distance, Manhattan distance, Tchebychev distance and Cosine distance to see which distance measurement method can work optimally on the k-Nearest Neighbor algorithm. The selection of k values also determines the results of k-Nearest Neighbor classification algorithm, so determining the k value also needs to be considered. The data used in this study is a dataset of cervical cancer. The highest accuracy results obtained using the Cosine distance measurement method that is equal to 92.559% with a value of k = 9. Based on the accuracy values that have been compared, the most optimal distance measurement method is Cosine distance with the best k value obtained is k = 9 even though this distance measurement method has the highest computing time which is equal to 0.898 seconds.