topic Re: Grouping days with similar distribution pattern in SAS Data Science

Grouping days with similar distribution pattern

kdp — Thu, 11 Jun 2009 20:42:21 GMT

Hello,

I have data by 15min intervals which indicates the percentage of demand that occurred during that 15min of the day. For ex:

06/11/09 8:00 - 2%
06/11/09 8:15 - 1.5%
06/11/09 8:30 - 5%
........
........

The percentages of all timestamps throughout the day add up to 100%. I have this data for one year.

What I want to do is group days with similar distribution patterns according to number of peaks and when those peaks occur.

The reason I want to do this is to figure out what driver is responsible for the different patterns.

Any ideas or thughts on how to approach the groupings?

Thanks!

Re: Grouping days with similar distribution pattern

DLing — Thu, 18 Jun 2009 13:15:54 GMT

How about clustering the days using the measurements as variables. This can put days with similar patterns together. Looking at the outcome clusters may give more insight on next step.

Re: Grouping days with similar distribution pattern

kdp — Thu, 18 Jun 2009 21:46:56 GMT

Thanks for the idea DLing!

I will try it out in Enterprise Miner and post my results.

Re: Grouping days with similar distribution pattern

kdp — Tue, 23 Jun 2009 16:17:05 GMT

DLing - your idea worked perfectly - Thank you once again!

I used one year of data as an input and the cluster node created four different clusters. Then I used the Segment Profile node to look at each cluster and see what driver values are represented in each cluster.

Although the clusters made sense, I wish it was more sensitive. It should have created anywhere from 10-20 clusters. Maybe there is some setting I need to adjust, so it becomes more sensitive and in turn creates more clusters.

kdp