I think you did a mistake by converting the original dataset into nominal, because this leads to make the titles for the features (3, 4, 5, 9) wrong.

1- I removed the titles for the above mentioned features and replaced with the word "Real" manually, using notepad++. Try to open this dataset (attached) and check the difference in weka. By using Real values for the features, weka will display (min, max std, mean ) for that feature. But, when you used nominal , weka deals with each number as a word.

2- I opened weka, opened the dataset, then I pressed on the button "Choose" in side the group of Filter (Preprocessing Tab).

3- Filter -> Unsupervised -> Attribute -> NumricToBinary (This is the second mistake you did).

4- Change the properties of the filter by pressing on the white textbox in the middle (Contains : NumericToBinary -R ) make sure the first textbox (attributeIndics is first-last) in order to convert all features.

5- Press on Apply ( next to the filter textbox).

6- save the processed dataset into a new arff file.

The resulted datasets is attached in my answer. However, I am not sure that was exactly what you wanted in your question. But I tried to explain how to convert the data from real values into binary directly.

Note 1: Features (12 & 13) didn't changed, because they are scaled already into (-1, 0, 1). I think -1 can not be changed into binary. You may correct these two features if necessary.

Note 2: There are several important filters you may check, such as NumericTransform, which helps you with converting the real values into integer values using floor function.

Note 3: Sometimes I prefer to prepare the dataset using a code written using a programming language. In your case, try to write your own code for converting the dataset into binary or whatever you want.

Hope this helps,

Regards ..

Using max-min normalization its possible to convert all data between 0 and one. This formula assign a=0 and b=1.

Sinan Salih Thank you very much I am in the middle of reproducing your steps. Would you mind elaborating on what you mean by replacing titles with the world "Real?"

Please check the attached files in the previous comment.

For more info, check the attached image in this comment.

Sinan Salih I understand now. Thank you very much I learned a lot.

Unless I am mistaken, I believe I cannot use this dataset for my intended purposes. Here is the link to where I received it from: https://data.mendeley.com/datasets/h3cgnj8hft/1.

My goal with this dataset was to train a decision tree to detect malicious URLs that are being passed to it by a flask API, ultimately informing the user that the URL is malicious.

The 13 attributes that are in the files I posted in my original post are what I found to be the most valuable of the original 48. The main issue I run into is that this dataset is numeric by default. I do not see how I can use this dataset to classify then inform the user. For example, this dataset is very popular and many projects on Github use it to accomplish something similar to what I intend to do: https://archive.ics.uci.edu/ml/datasets/phishing+websites. I believe it is popular because the attributes come binarized, it is already suitably formatted to start working with. Additionally, the meaning of each attribute is well-documented. I have yet to find an application that uses the 48 attribute dataset which further leads me to believe it is simply unsuitable to use for such an occasion.

I would love to use the dataset that I originally linked because it is newer and seems more reliable due to the methods used to gather the 48 attributes. Perhaps I am incorrect in believing this dataset can't be used for my purposes. Lastly, it seems that changing the dataset to binary really messes up the attributes given the nature of the values.

You are welcome

I think you have answered yourself in the last line.. what i did yesterday was just to help with the problem of converting or transforming.

However, transforming the data from a type into another is not a joke or that easy. Especially when transforming to binary values.

Maybe we can discuss more on using the new dataset and solve the problem using different ideas once i understand what are you planing to do...

Regards ..

John Smith

Having nominal data will not help you to achieve 0 and/or 1 label. To solve your problem, you need to have your data in a numeric type. Then, use "weka.filters.unsupervised.attribute.Normalize" filter. After that, apply the "MathExpression" filter where you need to define the right mathematical expression. In this case, use the expression: rint(A). The 'A' letter refers to the attribute being processed. Other attribute values can be accessed through the variables A1, A2, etc.

HTH.

Dr. Samer Sarsam