Performance of distance-based k-nearest neighbor classification method using local mean vector and harmonic distance
Dedi Candro Parulian Sinaga, Tulus Tulus, Poltak Sihombing · IOP Conference Series Materials Science and Engineering · 2020
Abstract K-Nearest Neighbor was one of the top ten algorithms data mining in the classification process. The low accuracy results in the K-Nearest Neighbor classification method was caused of this method used the system of majority vote which allowed the selection of outliers as the closest neighbors and in the distance model used as a method of determining similarity between data. In this process it is evident that local mean vector and harmonic distance can improve accuracy, where the highest increase in average accuracy obtained in the set data wine is equal to 6.29% and the highest accuracy increase for LMKNN is obtained in set data glass identification which is 16.18%. Based on the tests that had been conducted on all data sets used, it could be seen that the proposed method was able to provide a better value of accuracy than the value of accuracy produced by traditional K-Nearest Neighbor and LMKNN.