A Unifying View on Dataset Shift in Classification¶
Moreno-Torres, J. G., Raeder, T., Alaiz-Rodríguez, R., Chawla, N. V., & Herrera, F. (2012). A Unifying View on Dataset Shift in Classification. Pattern Recognition, 45(1), 521-530.
Cited by¶
5 citations across 5 artifacts.
Each citation links to the sentence it supports in the citing article.
Primes¶
- Concept Drift
- Consider a deployed binary classifier \(f: X \to \{0,1\}\) trained on data drawn from a joint distribution \(P_0(x, y)\), decomposed as \(P_0(x)\,P_0(y \mid x)\).
This sourceFormalizes dataset shift via the joint P(x,y) = P(x)P(y|x), separating covariate shift, prior-probability shift, and concept shift.
- Consider a deployed binary classifier \(f: X \to \{0,1\}\) trained on data drawn from a joint distribution \(P_0(x, y)\), decomposed as \(P_0(x)\,P_0(y \mid x)\).
- Data Drift
- The deployment distribution drifts — input statistics shift (covariate or feature drift), the input-output relationship shifts (concept drift), or the output base rate shifts (label drift).
This sourceTaxonomizes dataset shift into covariate shift, prior-probability (label) shift, and concept shift — the three axes along which the deployment distribution drifts.
- The deployment distribution drifts — input statistics shift (covariate or feature drift), the input-output relationship shifts (concept drift), or the output base rate shifts (label drift).
Domain-specific¶
- Ground-Truth Drift
- Label Shift
- Label shift names one specific subcase and prescribes specific corrective machinery that is distinct from and often cheaper than the corrections required for the other subcases
This sourceThat prior-probability (label) shift is one named cell of the dataset-shift taxonomy, distinct from covariate and concept shift and calling for its own correction.
- Label shift names one specific subcase and prescribes specific corrective machinery that is distinct from and often cheaper than the corrections required for the other subcases
Mechanisms¶
- Decision Tree Pruning
- It misleads when the holdout is not representative of where the model will be used — prune against a validation set drawn from the wrong distribution and you optimize for the wrong world, cutting branches that would have mattered in production.
This sourceDefines dataset shift as a mismatch between development data and the distribution encountered in unseen real-world use.
- It misleads when the holdout is not representative of where the model will be used — prune against a validation set drawn from the wrong distribution and you optimize for the wrong world, cutting branches that would have mattered in production.
Verification¶
Does it exist? Confirmed. This work's DOI resolves to a registered record, which fixes its identity. That is all it fixes.
Does it back the claim? Not recorded. Neither this nor any other of the 5 citations of this work carries a recorded support check.
Support is checked per citation rather than per work — the same source can be cited soundly in one article and wrongly in another. Per-citation recording began recently, so a citation with no recorded check is a gap in the record rather than evidence it went unchecked.
See how references were verified.
Registry ID ref:0af84f338d53 · see in the full table