cluster analysis

[fengyuwuzu]

I tried to run cluster analysis using the following code,but in the work.tree data, some of the ID (DUPI) were replaced with blanks.

proc aceclus data=cluster.data_cluster_with_trajactory out=Ace p=.03 noprint;
	var  betting_days count_game_types mean_wager sites_wagered sum_wager total_bet_times total_times_over_days ;
run;

ods graphics on;

proc cluster data=Ace method=ward ccc pseudo print=15 out=tree
   plots=den(height=rsq);
   id DUPI;
   var can1-can7;
run;

ods graphics off;

I also got the warnings as below:

WARNING: Ties for minimum distance between clusters have been detected at 37734 level(s) in the cluster history.
WARNING: The MAXPOINTS option value 200 is less than the number of clusters (44887). This may result in a dendrogram that
         is difficult to read. The dendrogram will not be displayed. You can use the PLOTS(MAXPOINTS=) option in the PROC
         CLUSTER statement to change this maximum.
NOTE: The data set WORK.TREE has 89773 observations and 21 variables.

any hints and suggestions? Thank you.

[stat_sas]

As most of the distances are equal to 0 resulting 37K levels in cluster history table. Seems like its hard to accommodate this number for tree.

[fengyuwuzu]

All the variables are heavily positively skewed. I will do log transform and try again. Maybe this can make somem difference

Maybe I should use fastclus, which is for k-means clustering, and Cluster is for hierarchical clustering. Am I correct?

[stat_sas]

Transformation may not solve the problem. A quick check for collinearity may be helpful to avoid including correlated variables in the analysis.

[fengyuwuzu]

Thank you. Indeed they are highly correlated.