1. Distributions & Cutoff
N = 1,000
Class Separation (d')
1.80
Prevalence (% Positive)
20%
Decision Threshold (τ)
0.50
Score Density P(s | Class)
Drag canvas to shift τ
Negatives
(0)
Positives
(1)
Cutoff τ
Confusion Matrix
Real-Time Counts
| Actual \ Pred | Pred (+) | Pred (-) | Total |
|---|---|---|---|
| Actual (+) |
TP
171
True Hit
|
FN
29
Miss (Type II)
|
200 |
| Actual (-) |
FP
53
False Alarm (I)
|
TN
747
Clean Rejection
|
800 |
2. ROC Space (TPR vs FPR)
AUROC =
0.925
ROC Curve
Random Guess (0.50)
Operating
Point (τ)
ROC Properties:
Evaluates threshold-invariant discrimination. Unaffected by changes in class prevalence.
Operating: FPR = 6.6% | TPR = 85.5%
3. PR Space (Precision vs Recall)
AUPRC =
0.792
PR Curve
Baseline (π)
Operating
Point
Metric Scoreboard (τ = 0.50)
Live Evaluation
Accuracy
91.8%
Precision (PPV)
76.3%
Recall (TPR)
85.5%
F1-Score
0.81
Operational Trade-off Analysis:
At threshold 0.50, the model catches 85.5% of positive events while keeping false alarms
to 6.6% of negatives.