WINP: a window-based incremental and parallel clustering algorithm for very large databases

We introduce a new clustering algorithm called WINP for very large databases. Two different sizes of handling objects were used in WINP to acquire high accuracy and efficiency. WINP creates a window to detect approximate locations of clusters before accurate clustering processing. Clustering on these locations will reduce a lot of computations and get a good performance. WINP is the first algorithm to realize both incremental clustering and distributed parallel clustering. The advantages of our new approach are: (1) it is very efficient; (2) it realizes distributed parallel processing and can be run on a number of workstations connected via local area network; (3) it introduces a novel incremental clustering method for new coming data in an already processed database; (4) it is effective in discovering clusters of arbitrary shape; (5) it is not sensitive to noise; and (6) it has some ability to deal with high dimensional points.