K-means & ISODATA Clustering Method

The K-means method and ISODATA Method are two basic clustering methods. As the name suggests, K-means to specify K classes, and then get the last K centers through the initial center iteration. The initial center can be selected randomly or randomly, or the first K samples can be taken as the initial center. The final result of the cluster is closely related to the initial cluster center. Different initial centers may have different results. One way to solve the initial value problem is the K-means ++ method, which will not be explained here. If you are interested, refer to the last few links in this article.

One disadvantage of K-means is that you must specify the number of clusters, which is sometimes not feasible. Therefore, it is required that the number of this category can be changed, which forms the ISODATA Method. By setting some classification splitting and merging conditions, the number of classes is automatically increased or decreased during clustering. Of course, this also brings about a problem, that is, this condition is sometimes not so good. Of course, in many cases, ISODATA can still produce reliable results.

The following figure compares the two clustering methods:

This article is original and reprintedArticleOr please note the picture, this article address: http://blog.csdn.net/houston11235/article/details/8511379

K-means

ISODATA

The above two figures show some differences. My personal feeling is that the initial cluster center may be better if it is close to the center of the entire dataset. In addition, this clustering is greatly affected by the initial center. To obtain stable results, you 'd better run it several times and classify the data in a category into a class.

Reference link:

1. K-means algorithm

2. K-means ++

3. K-means Wiki

4. K-means Matlab code

5. ISODATA Matlab code

6. Another ISODATA's MATLAB implementation, can hold n dimen1_data