Decision tree using Auto model

Arupriya_Sen

I am doing Fraud Detection Analysis with Decision Tree using Auto Model in RapidMiner version 9.3.0. The dataset and screenshots are attached below. Instead of getting a nice tree, I am getting just a single good leaf. This means that the Auto Model is predicting all the examples of the label attribute to be good.  they are not even showing the decision options. How do I get  a nice proper tree? can anyone help me with this?

varunm1

I did check the data and see that your tree is pruned a lot. Please see below screenshots for comparison. To get to the model in automodel, see below screenshot. You need to select model and click open process. Play with pruning parameters and see the results of decision tree model as shown below.



With Pruning (Default in Automodel): Performance AUC 0.5 



Without pruning (need to remove manually). Performance AUC 0.59, but the tree is big.



Hope this helps.

Varun

varunm1

Hello @Arupriya_Sen

Are you talking about maximal depth? If so, to manually change that you need to deselect Automatically optimize in auto model before selecting RUN. This is setting the tree depth to 25.

varunm1

Auto model splits data into 60:40 ration amd use 60 percent for training and 40 percent for testing, this is the reason you dont see all the data in confusion matrix as only 40% is tested.

IngoRM

the Auto Model is not even taking into  consideration all the 999 examples. Isn't that wrong? The True bad are 23 and 65 and the true good are 17 and 180.

The predictions you are seeing (in the Prediction tab as well as in the confusion matrix under Performance) are for the 40% hold-out set.  Please search the community for more details on this, this has been discussed before multiple times actually ;-)

Can you also help me with the Predictions of the decision tree under the Prediction Tab? there are various colours used which I can't interpret.

Green means that this value is supporting the prediction of this row, red means that a value is contradicting the prediction of the row.  The darker the color, the stronger the support / contradiction is.

Hope this helps,
Ingo

varunm1

Hello @Arupriya_Sen

One thing you should be careful in analyzing the predictions tab is that the supporting and contradicting are related to prediction column only. It means that a strong supporting predictor for a wrong prediction is actually bad and vice versa.

Hope this helps.

varunm1

Hello @Arupriya_Sen

The first screenshot related to date and time error is because the input attribute that is going into the Extract day of month operator is not in date time format. You can check this by double-clicking your data and then check in statistics if the attribute is of type Date or not. If they are either numerical or nominal type, then you can use "Numerical to Date" or "Nominal to Date" operators to convert them into date type.

The second screenshot is a warning and nothing more. Here is relevant discussion related to that.

https://community.rapidminer.com/discussion/12465/parameter-repository-entry-accesses-a-repository-by-name-wesseldoc-data1

The third screenshot, I didn't understand as I can't see an error. If you just round off warning symbols its really tough for us understand the issue. Run it and check the log, you can access log from VIEW --> SHOW PANEL --> LOG.

varunm1

Hello @Arupriya_Sen

Due to your status level on the community, we are unable to see the screenshots. @sgenzer can help you post screenshots.

Coming to the auto model question, did you check the process to see if the decision tree parameter pruning is selected. If so, it might be the reason your tree is pruned a lot and some time all. To open the process you need to select a model under a decision tree in the left bar and then click open process on the top.

Also please do not post duplicate questions on the community, they might get tagged as duplicate. You can modify your question using an edit option (Click on wheel icon and edit).

https://community.rapidminer.com/discussion/55732/decision-tree-using-auto-model#latest

Thanks
Varun

Arupriya_Sen

Thank you so much for the solution. It really helped. Can you tell me how to reduce the maximal gain parameter to 15 from 25? when I'm changing it and running the process, it is automatically set back to 25.

Arupriya_Sen

Thank you so much once again.  If I have to closely look at the Confusion Matrix or the performance of the tree, the Auto Model is not even taking into  consideration all the 999 examples. Isn't that wrong? The True bad are 23 and 65 and the true good are 17 and 180. 

Arupriya_Sen

Can you also help me with the Predictions of the decision tree under the Prediction Tab? there are various colours used which I can't interpret.

Arupriya_Sen

Can you help me in removing this error in the screenshot attached below? I have circled the error with red.

Arupriya_Sen

Can you help me remove the errors as well? The errors are circled in red in the screenshots attached below.
 

Arupriya_Sen

I am facing the following errors:


The 4th and 1st ones are for the same operators. The rest of the two are for different operators. 

Arupriya_Sen

Hello @varunm1, could you help me here? Thanks in advance.

Arupriya_Sen

@sgenzer any suggestions here? It's extremely urgent. 

sgenzer

@Arupriya_Sen hmm sorry I am completely lost on this thread. Can you please explain what's going on at this point?

Arupriya_Sen

@sgenzer I am doing Fraud Detection Analysis with Decision Tree using Auto Model in RapidMiner version 9.3.0. The dataset and xml process are attached below. Even though I am getting results, many of the operators are showing errors and I can't understand how to remove them. Could you please help me with the same? Urgent help needed. Thanks in advance. 

sgenzer

hi @Arupriya_Sen very sorry for the delay - I was on vacation from July 3 until now. Do you still need help with this?

Telcontar120

Many of these errors are not fatal errors but simply warnings.
For example  about the path locations of your datasets---if they are not set as relative paths, then if you move or share the process, the links will not work properly.
Or the error about the attribute name may be because you don't have any such attribute in the dataset, or the metadata for that attribute has not yet propagated.
In general, if your process runs and completes and gives you the expected output, then these types of warning errors can be disregarded.  Of course if your process doesn't complete, then you have a serious error that needs resolution.