Apply filters to Predictive intelligence model

Bharath38
Tera Guru

I have created a new Predictive intelligence model from workbench to predict Configuration Item for incoming incidents. We decided to apply this only for certain CI's so I have applied some filters to train on only those CI's. Net automation %79 which is good, however when we do batch test we get all incorrect predictions, I think it's because it's predicting other CI classes which we didn't train on. Can we apply filters to model to only predict for certain incoming incidents or it's has to all Incidents. Any help is appreciated !

 

Lener Pacania1  any insights please.

3 REPLIES 3

Lener Pacania1
ServiceNow Employee
ServiceNow Employee

Hi Bharath,

 

Your model will predict against all incoming incidents when implementing via PI workbench. If you want more control on when to apply the model you would build the classification model by going to all>Predictive Intelligence>classification and build and train your model there.  Then you would build conditional logic on when to fire that model via a business rule or flow designer. For example, let's say you trained a model to predict assignment group and you trained on your top 50 CIs.   You would build conditional logic that would only fire the PI model if the CI was 1-50 (assumes you know the CI) for everything else you would do something else.  Pseudo code would look something like IF CI=<50 then fire PI model ELSE use assignment rule.   Examples of business rule code to call your PI model is here.

 

If you are more comfortable with Flow Designer you can reference the example in our documentation here.

 

HTH - Lener

Thank you for reply.

 

I'm not seeing Predictive Intelligence spoke in our flow designer. I see there is one BR already for Sandiego release 'Default Task Based Prediction" when I activate it I got Error Message

applyPredictionForSolution(gr, solution) is only applicable for classification solution and Invalid insert.
 
BTW we are trying to predict Configuration Item on Incident based on the short description in turn we have a process that set Assignment group based on CI.

Hi @Lener Pacania1 , 

I was looking for something similar and came across this question on community. My question around this is if we use flow designer to predict the values conditionally and not use the PI workbench, Can we still configure and tune the Precision and coverage of our model ? Will tuning them in Pi workbench help at all as we are not using the template or is there any other way to tune these values ?