information value calculation in case of zero number of y class

[giorgos-kyriakopoulos]

Hi, iv_xy function replaces 0 with 0.9 in order to calculate information value. Would it be more appropriate to replace with a value closer to 0 , let's say 0.01?

[ShichenXie]

This replacement is to make sure the iv can be calculate in case of zero y labels. Replace with 0.1 or 0.9 would not affect the iv, since it might not occur in a real data sample after bining . I didn’t find any reference on this issue, if you have please post it here.

[giorgos-kyriakopoulos]

In a case that a binning process results to a bin with a small number of good class and zero of bad class then I suppose that replacing 0 to almost 1 (0.9) is different than replacing 0 to 0.01. What I mean is that in real cases where default rate is very low (2%) the fact that we appear defaults(0.9) in bins where the do not exist ( by converting 0 to 0.9 instead of 0.01) might lead to misconceptions.