Should i exclude the classes before i retrain Predictive Intelligence?

ACripps
Tera Contributor

3 weeks ago

Hi all,

The last time we retrained PI on our instances was a couple of years ago. Do I need to exclude the classes before I retrain PI ? 

Thanks

0 Helpfuls
  • 177 Views
1 REPLY 1

Itallo Brandão
Tera Guru

3 weeks ago

Hi ACripps,

The short answer is Yes, absolutely.

Since it has been a couple of years, your historical data likely suffers from "Concept Drift." If you retrain on the full dataset without exclusions, the model will learn from outdated patterns and may predict "Ghost Classes" (Assignment Groups or Categories that are retired or no longer in use).

Best Practices for Retraining after a long gap:

1. Filter by Time (The "Natural" Exclusion) Instead of manually picking which classes to exclude, tighten your Window of Time in the Solution Definition filter.

  • Recommendation: Set the filter to Created on > Last 12 Months (or 6 months, depending on your volume).

  • Why: This automatically excludes any classes that haven't been used in the last year, ensuring the model only learns your current business process.

2. Hard Exclusion (Active Check) Update your Solution Definition filter to explicitly exclude retired targets.

  • Example: If predicting Assignment Group, add: AND Assignment Group.Active | is | true.

  • This prevents the model from suggesting groups that still exist in the system but shouldn't receive new tickets.

3. Coverage vs. Precision Check your "Minimum records per class" property (usually defaults to 30 or 50). If you have classes with very low volume in the last year, excluding them (or letting the system ignore them via this threshold) increases the overall precision of the model.

If this response helps you solve the issue, please mark it as Accepted Solution.
This helps the community grow and assists others in finding valid answers faster.

Best regards,
Brandão.

0 Helpfuls