Predictive Intelligence Classification

[Rohit Singh]

I am trying to Implement Predictive Intelligence Classification where my output filed is Assignment Group and input filed is Short description. I have trained my solution model for 40K records. Now I am testing my ML solution however I am not getting how the system is going to predict Assignment group based on the historical data. There are various parameters Estimated Precision, Estimated Recall and Estimated Coverage which I am not getting how it is calculated and how is it playing role while predicting. By distribution I understand that it is the percentage calculated from the total number of incident assigned to a group divided by Total number of historical Incidents.   

Please help in providing clarity. Thanks in advance.

[Aniket Chavan]

Hello @Rohit Singh ,

Here's some info that might help you to understand it better:

1. How Predictive Intelligence Works
- Input Field: Short Description (key info on the incident)
- Output Field: Assignment Group (what the model predicts)
- The model uses historical data patterns from your 40K records, mapping terms in descriptions to groups.

2. Key Metrics Explained
- Estimated Precision: Shows how often the predicted group is correct. Higher precision means fewer incorrect assignments.
- Estimated Recall: Measures the model’s ability to capture all relevant groups. This is important for capturing all correct assignments.
- Estimated Coverage: Reflects the percentage of records the model can predict confidently. Higher coverage means fewer manual assignments but may slightly affect precision.

3. Data Distribution
- The distribution metric shows the frequency of each Assignment Group in historical data. For example, if Group A has 25% distribution, then 25% of incidents were historically assigned to it. This helps the model learn to predict common groups more accurately.

4. Training and Testing
- Training: Uses historical incidents to learn the link between Short Description and Assignment Group.
- Testing: Evaluates metrics like precision and recall. Visualize these metrics in the Solution Visualization tab, where higher precision and coverage groups appear in the top right.

5. Tuning the Model
- Balance Precision and Coverage based on your needs. High precision but low coverage may indicate conservative predictions.
- Consider Excluding groups with low distribution or accuracy to improve overall performance.

Please Mark ✅Correct if this solves your query and also mark 👍Helpful if you find my response worthy based on the impact.

Regards,
Aniket