Because the univariate performance of the predictor is below the threshold configured in the Adaptive Model rule.
Because the predictor is strongly correlated to another predictor. From such a group of strongly correlated predictors, only the best one is used.
Please note that predictors can change state (between active and inactive) during their lifetime. As new relations are "discovered" by the model, or the data changes, or the behavior of the population changes, the predictors are regularly reviewed and can become active or inactive.
The thresholds are not usually changed - you really need to know what you're doing. But if you have off-line data available you can use Predictive Analytics Director to get immediate insights in the effects of the thresholds mentioned.
Posted: 6 years ago
Posted: 24 Oct 2017 18:07 EDT
Parthiban Umamaheswaran (ParthibanU)
Because the univariate performance of the predictor is below the threshold configured in the Adaptive Model rule. Ans: In Adaptive model ,performance threshold is 0.52.
Because the predictor is strongly correlated to another predictor. From such a group of strongly correlated predictors, only the best one is used.Q:How to find it is correlated or not?
3. Q:I am practising Pega Decision Consultant course and I am not clear with predictors concept in Adaptive Models. Please explain how the predictors will be determined as active or inactive ? and as you said if it is based on threshold setting, please elaborate how can we monitor the threshold for each model ?
Note: I am practising in cloud environment provided at the end of course and not a real time one
Posted: 6 years ago
Posted: 30 Oct 2017 7:06 EDT
Otto Perdeck (Otto_Perdeck)
Director, Data Science, Machine Learning & AI
I'm afraid I don't understand how your question 3 is different from the earlier questions. Please elaborate on what's not clear about the explanation how predictors are becoming active/inactive.
Regarding "how to monitor the threshold" - that's not what you do. The threshold is a setting of the rule. What you would want to monitor is the actual performance of the models & predictors. The overview of Adaptive Analytics on PDN helps to understand the various tools that are available to do so, see https://pdn.pega.com/adaptive-analytics.
If you're looking for a generic into in ML or the concept the concept of "predictors", you could consider some generic textbooks or courses on the topic (e.g. check Coursera).