Track classification prediction results over time

  • Release version: Washingtondc
  • Updated July 23, 2024
  • 1 minute to read

    • Use the Prediction Results dashboard to determine if classification solution predictions are improving over time. Identify solutions that need refining or retraining.

    Before you begin

    • Role required: admin, ml_admin, or ml_report_user

    About this task

    The Prediction Results dashboard reports on coverage, precision, and recall over time for classification solutions.

    Statistics are provided in two timeframes: the average for the past 30 days, and daily. The indicators coverage, precision, and recall are defined as follows.
    Table 1. Prediction Results indicators
    Report type Definition
    Coverage The percentage of predictions that yielded an outcome out of the total number of predictions that were attempted.
    Precision The percentage of predictions where the predicted value was the same as the final value of the field when the report closed.
    Recall The percentage of correct predictions that yielded an outcome out of the total number of predictions that were attempted.

    Procedure

    1. Navigate to All > Predictive Intelligence > Classification > Prediction Results Report.
    2. In the Prediction Results dashboard Filter by solution prompt, select the solution statistics you want to review.
      The system updates the dashboard based on the solution you selected.
    3. Identify classes with anomalous coverage, precision, or recall values.
      For example, identify solutions where coverage, precision, or recall is declining over time.

    What to do next

    Refine the solution definition filter by including or excluding classes as needed. After updating, retrain the solution.