Concept Drift Detection Calculator

Work out concept drift detection instantly with clear inputs, formula shown and shareable results.

About Concept Drift Detection Calculator

The Drift Detection Method treats the error rate as a Bernoulli process and derives control limits from its standard error, sqrt(p(1-p)/n). Crossing two standard errors raises a warning and three signals concept drift — the relationship between inputs and target has genuinely changed. Because the limits shrink as the window grows, larger windows detect smaller shifts but respond more slowly.

Formula

DDM control limits

sigma = sqrt(p(1-p)/n); warning = p + 2 sigma; drift = p + 3 sigma

Worked example

With the default values below, Concept Drift Detection Calculator returns 3-sigma drift limit of 9.820%. Change any input above and every figure here is recalculated in your browser.

Inputs used

Baseline error rateaccepts 0.01–99.9 %: 8 %
Current error rateaccepts 0.01–99.9 %: 13 %
Observations in the current windowminimum 10: 2,000

Results

3-sigma drift limit: 9.820%
2-sigma warning limit: 9.213%
Error increase: 5.000%
Detector state: Drift - error exceeds the 3-sigma control limit

Step by step

1Values used
2DDM control limits
33-sigma drift limit= 9.820
42-sigma warning limit= 9.213
5Error increase= 5.000
6Detector state= Drift - error exceeds the 3-sigma control limit

Assumptions and limits

The result is derived only from the 3 inputs listed above. Anything not among them is not part of the calculation.
In particular, this calculator does not take inflation, fees or charges or compounding frequency into account.
Figures are rounded for display to 3 decimal places, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

How does concept drift differ from data drift?

Data drift is a change in the input distribution; concept drift is a change in the mapping from inputs to the target. Only concept drift necessarily degrades accuracy.

What should happen when drift fires?

Collect recent labelled data, evaluate the current model on it, then retrain on a window that emphasises the new regime. Alerting without a retraining path just produces ignored pages.