Question

8
Replies
86
Views
Close popover
Nanjundan Chinnasamy (Nanjundan Chinnasamy)
Verizon Communications
PEGA Lead Decisioning Architect
Verizon Communications
IN
Nanjundan Chinnasamy Member since 2015 9 posts
Verizon Communications
Posted: February 3, 2021
Last activity: February 19, 2021

ADM - Symbolic predictor binning issue

Hi Team,

We have ADM implementation in one of our usecase. We have seen our ADM is not performing compare to the business rules for certain use cases and when we debug the root cause we found a weird issue associated with how symbolic predictors are getting Binned impacting the ADM prediction. For example: a predictor "Reason" has 3 Bin's as per the attached screen print.  Bin1 : REFAN, SRLTE, SERREP, TERMSFEES Bin2 : MISSING Bin3 : NON-MISSING

From the example it classified Missing and Non-Missing as key values for predictors rather than Binning the other sets of values in Bin1 (we are expecting higher response for Bin1).

Current ADM Settings:

Data analysis binning

  • Grouping granularity: 0.25
  • Grouping minimum cases: 0.05

Predictor selection

  • Activate predictors with a performance above:0.52
  • Group predictors with a correlation above:0.8

We are expecting each symbol to be kept in a separate Bin to make the grouping of symbolic predictor more meaningful. Anyone come across the similar issue & what are the additional settings needed? We are planning to update "Grouping granularity from 0.25 to 0.9 and "Group predictors with a correlation above" from 0.8 to 0.95

***Edited by Moderator: Pooja Gadige to update INC details***
Pega Customer Decision Hub 8.3 Decision Management Communications and Media Consumer Services Decisioning Consultant Support Case Created