1 Reply Latest reply: Nov 1, 2006 7:32 AM by 265939 RSS

    O Cluster Algorithm giving only one cluster

      I am using Oracle 10g O Cluster Algorithm for clustering. i have 790 rows of data with 18 attributes all are of float data type.

      I am using ODM for creating clustering, but clustering resulting in creation of only one cluster.

      can any body knows what is the actual problem.

      please reply soon..

      thanks in advance
        • 1. Re: O Cluster Algorithm giving only one cluster
          It is possible for OCluster (OC) to return a single cluster if it did not find any natural clusters in the data. That is, it did not find any split in the data that would separate well the data into two groups. The number of clusters settings in OC is not the desired number of cluster but the maximum number of clusters. One way to validate this is to look at the histogram of your 18 attributes and see if there are clear peaks and valleys. In order for OC to find a split you should see at least two peaks with a reasonably deep valley (the actual criterion is statistical) in one of the attributes. If you don't see that than OC would not create more than one cluster. I am assuming that you are using ODMR and the data preparation steps suggested by ODMR.

          You can also try to increase the sensitivity parameter. This will allow OC to accept clusters that it would not consider significant otherwise.

          If you still want to partition your data even when OC did not find any clusters than you may try KM. KM will split the data into K clusters as long as there is enough data for that.