Operational Lift

Practical decision-support value of the model for triage and prioritization

Lift is useful for triage. It shows whether the highest-risk rows contain abnormal-demand events at a higher rate than the overall test base rate (~12.75%).

Lift by Risk Band

Top 1% predicted risk
4.49x
Lift over base rate
Precision0.57
True Positives1,418
Top 5% predicted risk
3.47x
Lift over base rate
Precision0.44
True Positives5,480
Top 10% predicted risk
2.97x
Lift over base rate
Precision0.38
True Positives9,364
All category-threshold alerts
2.36x
Lift over base rate
Precision0.30
True Positives16,247

Lift Over Base Rate by Risk Band

Detailed Lift Metrics

Risk SetRows / AlertsTrue PositivesPrecisionRecallLiftBase Rate
Top 1% predicted risk2,4761,4180.57270.04494.49x12.75%
Top 5% predicted risk12,3805,4800.44260.17363.47x12.75%
Top 10% predicted risk24,7599,3640.37820.29662.97x12.75%
All category-threshold alerts53,94116,2470.30120.51462.36x12.75%

Precision by Risk Band

How much better is focusing on top-risk rows vs. random selection?