Uncertainty and models

Precision-recall curve

Shows precision against recall across binary decision thresholds.

Plot
Model evaluation
Threshold comparison

What it answers

How does precision change as recall increases?

Shape: Precision versus recall curve
What it shows: Retrieval trade-offs and false-positive burden.

Example

The question

How many flagged customers really churn as we flag more of them?

Churn model: precision against recall (PR curve)

As recall rises, precision falls: catching half of leavers gives 70% correct flags; catching 90% drops precision to 30%. The dashed line is the 18% base rate.

Illustrative data. It belongs to no company.

When to use it

Use it when

Binary evaluation where positive-class retrieval matters.

Avoid it when

Comparing populations with different prevalence without context.

How to read it

Read recall on X and precision on Y while thresholds change.

Note

Precision depends on prevalence.

The data you need

Layout: Long table
Fields: CaseID: text; Actual: 0/1; Score: number
One row is: One case with its true binary outcome and prediction score. What is the grain?
Rules: Define the positive class and score direction. Use held-out predictions for evaluation. Calibration needs probabilities in [0,1].
Optional: Weight. ROC/PR alternatives: precomputed Threshold,TP,FP,TN,FN with complete threshold coverage.
Preparation: Use held-out binary outcomes and scores; calculate precision and recall at the selected thresholds.
Units: Precision=TP/(TP+FP); recall=TP/(TP+FN). State zero-denominator handling and positive-class prevalence.

A long table: one row per observation, few columns. New data adds rows.

Worked example

Positives 0.9,0.65; negatives 0.55,0.2. Threshold 0.6 givesprecision 1, recall 1; threshold 0.5 givesprecision 2/3, recall 1.

CaseID	Actual	Score
A	1	0.90
B	0	0.20
C	1	0.65
D	0	0.55

All charts in the library

Analytics cookies