Comparing Rapidminer decision tree and Weka's

Pirehelo

Hi everybody 

Can someone please explain how Rapidminer decision tree operator is different from J48 (or W-J48) decision tree in Weka? The accuracy I get from the latter is considerably higher. Weka's documentation clearly mentioned that J48 is based on C4.5 algorithm. How about Rapidminer's? If they are the same why do they give different accuracies for similar parameters? Plus, despite Rapidminer's, Weka's cannot handle missing data point. 

 

Thank you 

sgenzer

hello @Pirehelokan - did you search for this thread?

 

https://community.rapidminer.com/t5/RapidMiner-Studio-Forum/Decision-Tree-Operator-C5-0-or-C4-5-algorithm/m-p/31187#M22488

 

Scott

 

 

sgenzer

hello @Pirehelokan - did you search for this thread?

 

https://community.rapidminer.com/t5/RapidMiner-Studio-Forum/Decision-Tree-Operator-C5-0-or-C4-5-algorithm/m-p/31187#M22488

 

Scott

 

 

Pirehelo

Thank you Scott,

 

That thread was pretty helpful. However, I could not see the source code of the decision tree. I understand that rapidminer decision tree could use both information gain and gini index for splitting nodes. But even when I use the information gain its accuracy is different from Weka's C4.5. So, I was curious to see how is the tree implemented?

 

Thanks,  

 

sgenzer

so the sourcecode for the entire RapidMiner Studio Core is on GitHub. Have at it.

 

Scott