The Cluster node in SAS^® Enterprise Miner™ enables you to cluster large data sets quickly. The clusters are formed using k-means methods. Sometimes the node produces one large cluster (some say "no clusters" in this context) with a few very small clusters. In this scenario, the following causes might be relevant.

• Your data contains only one large cluster, and the node correctly identified it. Having a single cluster is a mathematical possibility. If you want more clusters, consider using more variables, or fewer variables.

• The data needs to be standardized. By default, the variable values are divided by the standard deviation. The mean is not subtracted. Variables with large variances tend to have more effect on the resulting clusters than variables with small variances. If all variables are measured in the same units, then standardization might not be necessary. Otherwise, some form of standardization is strongly recommended.

• The clusters are irregularly shaped. K-means methods are very good at finding clusters that are well-separated, compact, and spherical. If your clusters are substantially different, then consider nonparametric methods such as those in PROC MODECLUS in SAS/STAT^® software.

• The data contains outliers. The Filter node enables you to apply a filter to the data set in order to exclude outliers or observations that you want excluded from your data mining analysis.

• The wrong number of clusters is selected. The CCC plot does not give a clear indication of the number of clusters. To interpret the plot, see the chapter "Cubic Clustering Criterion" in SAS Enterprise Miner Help. Look for the following warning in the Cluster node log:

  WARNING: The number of clusters selected based on the CCC values may not be valid.  
  Please refer to the documentation on the Cubic Clustering Criterion.
  

The list of causes above is common in practice, but the list is not exhaustive. Further analysis of your data might be required. The following references contain additional information about cluster analysis.

• "SAS/STAT User's Guide" in SAS Product Documentation: refer to the clustering introductory chapter, as well as chapters for the CLUSTER, FASTCLUS, and MODECLUS procedures.

• SAS Enterprise Miner Help: from within SAS Enterprise Miner, select Help ► Contents to access the Help. Customers who license SAS Enterprise Miner can also access the Help from our website. SAS Enterprise Miner. To request User ID and Password information, contact SAS Technical Support and provide your SAS Site number.

Operating System and Release Information