CODEWITHSUNDEEP
classificationwekasvm
user1228310
user1228310

Reputation: 121

How to perform nominal to numeric conversion of attributes in WEKA?

I have a dataset containing a mixture of numerical and nominal attributes. I want to convert all the nominal attributes in the dataset to numeric so that I can apply the SVM classifier kernel(PolyKernel and RBFKernel) that only works with numeric attributes. Any help would be greatly appreciated. FYI I've already tried NominalToBinary Filter(Its not really what I want)

Upvotes: 1

Views: 17907

Answers (4)

Sakshi Maurya
Sakshi Maurya

Reputation: 51

Yes, you can convert nominal data to numeric in weka:

Example: enter image description here

select: filters.unsupervised.attribute.OrdinalToNumeric

enter image description here

Upvotes: 0

fracpete
fracpete

Reputation: 2608

The NominalToNumeric filter (package: weka.filters.unsupervised.attribute) that is part of ADAMS allows you to do exactly that. You can either use the internal representation (i.e., order of labels starting at 0), or, if there is a numeric part in the label that can be turned into a number, use regular expressions to convert these sub-strings.

ADAMS also offers the Weka Investigator, a more powerful tool than the Weka Explorer. Just download the adams-ml-app-snapshot snapshot to get access to this filter.

Upvotes: 0

Chandima Samarakoon
Chandima Samarakoon

Reputation: 409

There is no direct filter to convert nominal data to numeric data. If your nominal attribute has 2 values (SEX: MALE, FEMALE) you can easily apply the filter under unsupervised filters "nominal to binary".

But if you have more than 2 variations for the attribute, you cannot use "nominal to binary". So you need to use a filter called "Rename Nominal vales". There you can convert nominal value to numeric value. Eg: if your dataset has an attribute called " region" and it has "INNER_CITY, TOWN, SUBURBAN, SUBURBAN" for values, you can easily convert those nominal values using the "Rename Nominal Value" filter.

There is a value replacement form, you have to do only add values like below. INNER_CITY:0, TOWN:1, SUBURBAN:2, SUBURBAN:3

you can see your results.

Upvotes: 0

Matthew Spencer
Matthew Spencer

Reputation: 2295

One thing you could do is convert all of the label names for the attribute using RenameNominalValues. Please note that all of these new labels would need to be numeric, so you might need to change them as below:

enter image description here

Once this is done, then you could save the .ARFF File, and change the entry in your attributes list from:

@attribute a0 {false,true,maybe}

to

@attribute a0 numeric

Once saved, reload the document and hopefully all will load okay.

Alternatively, you could try your favorite Spreadsheet Application if conversion of your data back to ARFF would not be an issue.

Hope this Helps!

Upvotes: 3

Related Questions