"Attribute selection using clustering"

wessel

Dear All,

Is it possible to do attribute selection using a clustering algorithm?
The idea is, a clustering algorithm clusters instances,
but if you take the transpose of the dataset, the algorithm will cluster attributes.

The following procedure seems a little weird, so I would like to hear your opinion:
- load a dataset with 23 000 instances and 1400 attributes
- transpose this dataset (invert columns and rows)
- apply a clustering algorithm
- merge all "clustered attributes" into "new single attributes"

IngoRM

Hi,

this is indeed far from weird and might in some cases actually be a very good idea. Since the attributes (examples after the transposing) do not have to be selected but also could be aggregated into new attributes this approach can serve not only as attribute selection but also attribute aggregation approach.

In analysis of microarray data this approach is often used as far as I know. I am not sure if the data is there transposed also for attribute selection / aggregation by clustering or only for handling the problem that in microarray data usually much more attributes than examples exist.

Cheers,
Ingo