Minimum Observation Count for IV value?

[Ishankhare1]

Hi Thomas,

I am getting zero information value for many variables having low observation count. I was getting significant IV value using 'Information' package in R for these variables. What is the cutoff for minimum observation count to get non-zero IV value in this package? Is it minimum 5% observations in each bucket?

Regards,
Ishan

[tomasgreif]

What data do you have. In case it is heavily skewed it might be some issue. Or try to use different rpart control in case this is in fact using decision tree to make the splits.

[Ishankhare1]

Information Package Results:

Sr | Independent_Var_2 | N | Percent | WOE | IV
1 | [0,0] | 23921 | 0.946504174 | -0.654004424 | 0.300204339
2 | [0,1] | 1352 | 0.053495826 | 2.441050188 | 1.420707247

Tomasgreif Package Results:

Variable | InformationValue | Strength | class | outcome_0 | outcome_1 | woe | miv
Independent_Var_2 | 0 | Wery weak | (;) | 24738 | 535 | 0 | 0

As you see the IV from information package is 1.42 and from this package it is 0.00. I cant share actual data but you can populate as its binary. The total non zero independent var_2 is 5.3% so I was thinking is there a cutoff in this package for minimum observations in a bin?

Below is the data:

Dependent Var_1	Independent Var_2	Frequency
0	0	1083
1	0	269
0	1	23655
1	1	266

Kindly let me know if i should use the results of this package or refer to information package?

Ishan