Is it possible to configure Adaptive in such a way that it would limit # of active predictors it's using e.g. to top 15 predictors by AUC? From one hand AI behind Adaptive would be happy to consume lot of data points and build propensity scores based on it but from the perspective of GDPR data minimization I need to ask if it is possible to only consume the top performing-predictors per model.
The current active predictor threshold functionality is not helping as in case I'd set it to e.g. 0.65 then it could be that lot of models still continue to have huge amount of active predictors, while some low-performing models might lose even the slight intelligence it has.
Also decreasing the overall amount of predictors in Adaptive would not be good idea, as different models have different performances for different predictors. So, e.g. if would limit the total list to 100 predictors then it still might happen that some models continue to have significant amount of active predictors while other models might not have any relevant predictors in model any longer.
Just need to understand if such business request would technically be even possible to do or we need to pick one of the not-that-good solutions mentioned above.
There is no such option as it would seriously hamper the models. Why would you want to do this? What's driving this request? I'm not aware that GDPR drives "data minimization" - for what I know it is more about the use of certain fields/predictors but not necessarily the amount of them. Can you elaborate?
In a reporting layer you could of course just show the top-N predictors.
@Otto_Perdeck , thanks for quick answer. This question is not related to reporting but concretely about the sensitivity related to customer personal data processing for marketing purposes (what Adaptive with prioritizing our NBAs is actually doing, processing customer personal data). Even with consent from customers GDPR directive states that customer personal data should be used sparingly. Thus we have internal discussions whether we can continue to use Adaptive at all or use it, but with optimizing number of predictors per model. As I'm not aware about all ADM technical capabilities then thus I wanted to ask if such configuration (e.g. "15 active predictors per model") is even possible to configure in Pega? Either way there will be negative impact for the AI within Adaptive, the question is if we could apply the least worse option.
Otto_Perdeck , just to confirm and close this topic. It is technically not possible to limit # of Adaptive predictors per model and the only way we could limit is across models on whole Adaptive by removing some predictors if this is the legal request, correct?
But yes, from business perspective we are aware of the negative consequence on Adaptive performance if we would follow this road.
It's possible to configure the activation of predictors by setting a threshold for the AUC of predictors. In ADM, you can see this as a process of feature selection. However, considering the example you gave in your first question, currently, it is not possible to limit the number of adaptive predictors by setting the maximum number of predictors.
You can, of course, delete a predictor from the model rule, but it's not possible to limit the maximum number of predictors per model.