A Grid Based Clustering Algorithm
Qiang Zhang · 2010
To overcome the problems of Euclidean distance based clustering algorithms, an efficient algorithm CES is proposed. A distance metric derived from the infinite norm is introduced to measure similarities between objects, through the distance metric, the neighbor searching is converted to the intersection of projection sets searching, which speed up the clustering processing. An efficient neighbor searching method is proposed to improve clustering processing. A mathematical prove is given for the correctness of CES, and it is also proved that CES has the same accuracy as DBSCAN, but much faster than the latter. The theoretical analysis and performance experiments show that CES is effective in discovering clusters of arbitrary shape; it is very efficient with a complexity of O(N); it is robust against noise; it has a determinate result, not depending on the order of processing.