How to filter Predictive Intelligence PA metrics by specific ML Solution (Assignment group, etc)

[Brando008]

Hi everyone,

I am building a Performance Analytics dashboard to track our Predictive Intelligence models. I am looking at indicators like "Number of attempted predictions" and "Prediction Precision" which pull from the ml_predictor_results table.

Right now, the data is aggregating all of our ML models together. As you can see in the attached KPI Details screenshot, my results include ml_incident_assignment, ml_incident_categorization, and our custom subcategory models all mixed together into a single metric.

Goal: I want to be able to filter my dashboard widgets so I can see the metrics for only incident assignment, or only subcategories independently.

Question: How do I properly isolate these? Do I use an out-of-the-box breakdown (like Classification Solution) as a dashboard filter, create a new breakdown, or do I need to create completely separate Indicator Sources for each individual ML model?

Any guidance is appreciated!

 

[Vikram Reddy]

Hello @Brando008,

Use the out-of-the-box Classification Solution breakdown, don't build separate Indicator Sources per model. Looking at your Prediction Precision indicator's data source, the Available breakdowns field already lists Classification Solution and Predicted Class, both populated straight off ml_predictor_results (that's the Solution column you see in your KPI Details grid: ml_incident_assignment, ml_incident_categorization, and your custom subcategory model, sourced from ml_solution). Add that breakdown to the widget's Breakdown filter, or drive it from the dashboard's Filter by solution control you already have set up, and each widget will resolve to a single ML solution instead of blending them into one number.

References

• Predictive Intelligence concept (ServiceNow docs)
• Configuring advanced settings for ML solutions (ServiceNow docs)

Thank you,
Vikram Karety
Octigo Solutions INC