RapidMiner 9.7 is Now Available

Lots of amazing new improvements including true version control! Learn more about what's new here.

CLICK HERE TO DOWNLOAD

"Problem with decision tree algorithm"

szymekszymek Member Posts: 3 Contributor I
edited June 2019 in Help
hi,

I tried to run the Decision tree algorith in Raipd Miner and it seems not to provide a correct result. I am not sure if the problem is caused by the implementation of the algorith or there is another reason for that. Below is the exercise that I tried to run with RM.

I use the following data (A and B are nominal, binary attributes and there are two classes: + and-):
A,B,Class
T,F,+
T,T,+
T,T,+
T,F,-
T,T,+
F,F,-
F,F,-
F,F,-
T,T,-
T,F,-

I want to build a decision tree using Ginin index as the criterion for splitting. Rapid Miner selects attribute A as the best one for splitting. However, if I make calculations manually, B seems to be better. Do you know where is the difference from? Below are my calculations:
The overall gini before splitting is:
Gorig = 1βˆ’ 0.42 βˆ’ 0.62 = 0.48

The gain in gini after splitting on A is:
GA=T = 1βˆ’(4/7)2 βˆ’(3/7)2 = 0.4898
GA=F = 0
Ξ” = Gorig βˆ’ 7/10 GA=T βˆ’ 3/10 GA=F = 0.1371

The gain in gini after splitting on B is:
GB=T = 1βˆ’(1/4)2βˆ’ (3/4)2 = 0.3750
GB=F= 1 - (1/6)2 βˆ’ (5/6)2 = 0.2778
Ξ” = Gorig βˆ’ 4/10 GB=T βˆ’ 6/10 GB=F = 0.1633

Therefore, attribute B should be chosen to split the node (and not A as calculated by RM).

regards,
Szymon

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531   Unicorn
    Hi,
    thank you for this hint. We will check that, but might take some time.

    Greetings,
    Β  Sebastian
Sign In or Register to comment.