Predictive Intelligence recall definition

Gary Kakazu1
Mega Guru

Can someone give me a detailed definition of what recall means in Predictive Intelligence? I've gone over the docs, blogs, community postings and about the best thing I can find is: Percentage of false positives captured

Is that all there is too it? Identifying false positives?

1 ACCEPTED SOLUTION

Lener Pacania1
ServiceNow Employee
ServiceNow Employee

Hi Gary, sorry you had to wait 25d for response on this one.  Completely missed your question.

Recall is the ratio of True Positives/True Positives + False Negatives (see below graphic).

From a tuning perspective we increase recall to decrease false negatives and we increase precision to decrease false positives (see below graphic).  

What this means in predictive intelligence is that you have the ability to set the target for recall, which then adjusts precision and coverage dynamically to hit that recall target.  Same with precision.  This allows you to tune the PI model w/o having to add new inputs to your solution definition.  I created the below for one of my customers using CSM customer and it's based on a great LinkedIn article.   HTH - Lener

 

find_real_file.png

 

 

 

View solution in original post

4 REPLIES 4

Lener Pacania1
ServiceNow Employee
ServiceNow Employee

Hi Gary, sorry you had to wait 25d for response on this one.  Completely missed your question.

Recall is the ratio of True Positives/True Positives + False Negatives (see below graphic).

From a tuning perspective we increase recall to decrease false negatives and we increase precision to decrease false positives (see below graphic).  

What this means in predictive intelligence is that you have the ability to set the target for recall, which then adjusts precision and coverage dynamically to hit that recall target.  Same with precision.  This allows you to tune the PI model w/o having to add new inputs to your solution definition.  I created the below for one of my customers using CSM customer and it's based on a great LinkedIn article.   HTH - Lener

 

find_real_file.png

 

 

 

Thanks Lener, this helps.

Hi Lener,

 

So in the case of Incident Assignment group classification, how would this be defined?

You want the PI model to have high recall if it's disastrous for the PI model to predict incorrectly. For example, let's take a PI model that predicts the assignment group using short desc as the input. When an incident has a short description that reports a virus that could potentially breach the companies firewall we ALWAYS want the PI model to predict the network security team as the assignment group. We never want to assign those incidents to the wrong assignment group, because if we send it to the wrong assignment group it may sit there while the network is getting breached and the company is getting hacked.  In high recall situations it's ok for the network security team to get the occasional incident where it's not a threat to network security.   
 
HTH -Lener
PS: I welcome any of our product managers/customers/partners to join in with their examples.