Examining cause for mislabeling

b00122599

Hey folks,

I have completed a project and have an output for a decision tree predicting 4 classes. One of the classes I have a class precision of 99% then 69%, 66% and finally only 54% for my last class. What I want to do now is go through the validated example set outputted from the decision tree and examine which labels where incorrectly labeled, and what attributes are influencing the mislabeling. Is there any process in rapidminer that would be useful for this or would it be more of a manual job?

Thanks in advance,

Neil. 

varunm1

Hello @b00122599

You can use explain predictions operator to understand the impact of attributes on predictions.

You can go through the process in below link how to extract attributes that worked on correct predictions, if you change the filter you can also get wrong prediction related attributes.

https://community.rapidminer.com/discussion/54852/explain-predictions-ranking-attributes-that-supports-and-contradicts-correct-predictions#latest

varunm1

Hello @b00122599

You can use explain predictions operator to understand the impact of attributes on predictions.

You can go through the process in below link how to extract attributes that worked on correct predictions, if you change the filter you can also get wrong prediction related attributes.

https://community.rapidminer.com/discussion/54852/explain-predictions-ranking-attributes-that-supports-and-contradicts-correct-predictions#latest

b00122599

Thank you very much Varun!

Telcontar120

You can also filter your dataset for the incorrect predictions very easily if you want to profile those examples or examine them in more detail.  It is one of the built-in options in the Filter Examples operator (check the drop down parameter).