Weka Classifier does not accept string values

Question

I am using Weka to classify a data set. The .arrf data file looks like this. The problem I am facing while classifying is that A lot of classifiers like NaiveBayes etc. won't take the string attribute to classify it. Those two string attributes are important features for classification. I tried converting the string to nominal type using the filter but it doesn't convert it to nominal. How should I go about it considering the dataset I have ?

@RELATION transaction

@ATTRIBUTE transactionType  {'CC Credit',Trans,Exp,Dep,Check}
@ATTRIBUTE number numeric
@ATTRIBUTE posting {Yes,No}
@ATTRIBUTE String1 string
@ATTRIBUTE String2 string


@ATTRIBUTE amount real
@ATTRIBUTE class {1,2}


@DATA
'CC Credit',?,Yes,'XYZ Bank','ONLINE PYMT Aug',-1582100.38,1   
Trans,?,Yes,?,'ACH DEBIT XYZ CREDIT CRD-EPAY',-59219.40,2   
Exp,?,Yes,'First Nolastname','ACH DEBIT First Nolastname-RECEIVER',-176011.56,2

G5W G5W · Accepted Answer · 2018-09-07T23:40:47

You do not say what interface you are using. I assume that you are using the GUI.

On the "Preprocess" tab, under "Filters" select

filters -> unsupervised -> attribute

scroll down to find StringToNominal . By default, it will just convert the last attribute. You will want to change it to convert all of your strings.

Just in case:

If you are using R and RWeka, you can get this filter by running

Str2Nom = make_Weka_filter("weka/filters/unsupervised/attribute/StringToNominal")
Str2Nom(transactionType ~ ., data=Transaction, control=Weka_control(R=4:5))

Weka Classifier does not accept string values

2 Answers