K-Mean Algorithm with a Distance Based on the Characteristic of Differences
Zhong Li, Jinsha Yuan, Hong Ye Yang, Ke Zhang · 2008
A non-metric distance measure for similarity estimation based on the characteristic of differences is presented. This kind of distance is implemented in the well-known k-means clustering algorithm. To demonstrate the effectiveness of the distance we proposed, the performance of this kind of distance and the Euclidean and Manhattan distances were compared by clustering Iris dataset from the UCI repository. Experiment results show that the new distance measure can provide a more accurate feature model than the classical Euclidean and Manhattan distances.