Zeljko Vujovic

Supposed you loaded 4 datasets. All of them have a nominal class. However, the error you had indicates that not all these datasets have nominal class. Thus, you need to recheck your data and compare them with the following correct data that I have created and attached here.

HTH.

Dr. Samer Sarsam

Thank you very much.

I analyze detailed accuracy by class.

What are Feature Names, Feature Values and Discretization (Items) and how are they used?

The class is the last feature in your datasets. Since you are doing classification, the class must be from a nominal type. However, the error you had showed that at least one data (from the 4) did not have a nominal class. For this purpose, I created 4 datasets based on the tutorial you are following and asked you to compare your data with.

Dr. Samer Sarsam

I acted as you told me and resolved the problem - eliminated the error.

Now I have a new problem. How do I interpret the results: Detailed Accuracy By Class and Confusion matrix, that appear in the window Classifier output?

Zeljko Vujovic

In general, Feature Names are the names of your attributes/variables. However, Feature Values are the values of a variable belongs to each instance/example. and Discretization. The discretization in WEKA has 2 versions: supervised and unsupervised. The Discretize filter transforms numeric features into the nominal ones.

In order to interpret the result, you need to read about terms: TP Rate, FP Rate, Precision, Recall, F-Measure, MCC, ROC Area, and PRC Area. There is a huge amount of information in the literature, so you can read about each one.

Confusion matrix helps you to understand the overall performance of a machine learning algorithm. You need to look at its diagonal: the higher the values, the better the classifier.

For more information, you can refer to some of our papers like:

- The role of personality characteristics in informing our preference for visual presentation: An eye movement study

- Predicting user preferences of environment design: a perceptual mechanism of user interface customisation

Cheers,

Dr. Samer Sarsam

Samer Sarsam

Thank you very much for your thorough reply. I will review your paper.

My concrete tasks to make modul for Hepatitis C Virus For Egyptian Patients Data Set.

I have downloaded Discretized version of that dataset, from

https://archive.ics.uci.edu/ml/datasets/Hepatitis+C+Virus+%28HCV%29+for+Egyptian+patients (It is attached.)

There are 30 items and 3 attributes: Feature Names, Feature Values and Discretization(Items).

If I perform classification algorithm on taht dataset, with 10-fold cross validation, I am able to select: Feature Names, Feature Values or Discretization(Items).

What does it mean? What is the meaning of Feature Names, Feature Values and Discretization(Items) in this case?

Are clear my questions?

Sincerely yours,

Željko Vujović

Zeljko Vujovic

I have attached the original data before applying the Discretize filter. This is the data that you need to work with either with Discretization or without.

Discretize is not a compulsory task; it is just a technique to transfer the data from numeric to nominal type. The file you provided "Discretization-Criteria" showed the result after applying the Discretization filter on the data (which was done previously). To understand all the details you need to read the related paper(s) highlighted in the link you provide. Anyway, I have explained below the meaning of the terms (Feature Names, Feature Values, and Discretization) from my own understanding.

- Feature Names (name of the original attributes):

*Age *Gender *BMI Body Mass Index *Fever Fever *Nausea/Vomting Nausea/Vomting *Headache Headache *Diarrhea Diarrhea *Fatigue & generalized bone ache Fatigue & generalized bone ache *Jaundice Jaundice *Epigastric pain Epigastric pain *WBC White blood cell *RBC red blood cells *HGB Hemoglobin *Plat Platelets *AST 1 aspartate transaminase ratio *ALT 1 alanine transaminase ratio 1 week *ALT 4 alanine transaminase ratio 12 weeks *ALT 12 alanine transaminase ratio 4 weeks *ALT 24 alanine transaminase ratio 24 weeks *ALT 36 alanine transaminase ratio 36 weeks *ALT 48 alanine transaminase ratio 48 weeks *ALT after 24 w alanine transaminase ratio 24 weeks *RNA Base RNA Base *RNA 4 RNA 4 *RNA 12 RNA 12 *RNA EOT RNA end-of-treatment *RNA EF RNA Elongation Factor *Baseline histological Grading Baseline histological Grading *Baselinehistological staging Baselinehistological staging

- Feature Values (is the range Minimum-Maximum for each numeric feature).

- Discretization is the discretization value/range.

I recommend you to make your own settings and not to copy the settings/techniques of others. Discretization is not necessary to be a perfect choice; hence, you can try it and try other methods as well.

Cheers,

Dr. Samer Sarsam

How do I understand your recommendation to make my own settings without using the settings / techniques of others?

What does it mean that Discretization is not necessary to make the perfect choice? (Without discretion, I couldn't get my attributes to be nominal, which I needed.)

What other methods should I try, for example?

Zeljko Vujovic

Please find my answers below.

How do I understand your recommendation to make my own settings without using the settings / techniques of others?

Answer

Be creative. This is my point.

What does it mean that Discretization is not necessary to make the perfect choice? (Without discretion, I couldn't get my attributes to be nominal, which I needed.)

What other methods should I try, for example?

Answer

This depends on the goal of research.

Cheers,

Dr. Samer Sarsam

Samer Sarsam

Thanks for the replies. They are very helpful to me.

Let's look at the dataset that I need to model and determine the measures of its validity.

Original dataset: 1386 instances and 29 attributes.

The discretized data set was obtained from the original and has: 29 instances and 3 attributes. In so doing, instances of the discretionary dataset are attributes of the original dataset.

Questions:

1. what is a class in the original dataset?

2. what is a class in a discretized dataset?

Regards,

Željko Vujović

P.S. I have a problem loading the original dataset in WEKA and classifying it. what could be the reason for that?

Zeljko Vujovic

No problem. Please find my answer below.

Questions:

1. what is a class in the original dataset?

2. what is a class in a discretized dataset?

Answer

For any data-realted inquiries, you may need to contanct the researchres, who provided that, at: m.nasr '@' cis.asu.edu.eg

Note: I have attached the dataset in ARFF format yesters to make things easy for you. And I am also reattaching it here again. However, after contacting the authors, you will be able to know what is the class to predict.

Cheers,

Dr. Samer Sarsam

Samer Sarsam

It is very interesting and very important.

What can I do with these datasets in general if I don't know what to predict?

I got one opinion that, in the other case, I could choose from: Feature Names, Feature Values, and Discretization (Objects), which can be the output class. Is that possible?

Regards,

Željko Vujović