Hi Nanjundan,

I can't see the image you attached in the original post. A closer look at the ADM model reports would certainly be helpful. If that is something you rather not share publicly please do raise a support ticket with Pega and we take it from there.

But as per Sara's comment, the binning ADM creates reflects what is statistically significant. By forcing more bins by changing the threshold your models might start overfitting (as there is too little evidence for each individual bin). If there are a lot of distinct values in your symbolic fields you may want to consider making these less granular (e.g. instead of for example "Company name" use "Industry" - just as an example). The default settings of ADM usually do not need to be changed.

Note that ADM performance is also (heavily) effected by the way the responses are sent back. I don't know if there have been any changes to that, but if there is for example an ordering in the responses in some channel (e.g. all positives first) that has a detrimental effect on the models. If many values are missing that will of course also play a significant role.

Is the behaviour you are observing different now than it was before? Have there been any changes?

Thanks

Otto