SYSTEMA CONSTRUCTUM

Accepted ontology entry

leakage

Data leakage is a machine learning failure mode in which information from outside the training scope — test set data, future observations, or target-correlated features — enters the training process, producing artificially optimistic perfo…

ACCEPTED THINGcmsernc8k05tg3vv3qo17g356

Definition

Data leakage is a machine learning failure mode in which information from outside the training scope — test set data, future observations, or target-correlated features — enters the training process, producing artificially optimistic performance estimates. It arises from improper preprocessing ordering, insufficient temporal separation, or target variable leakage. The persistence mechanism is methodological documentation and validation protocols in machine learning practice. [formal: fuga | substrate: behavior | horizon: a moment | explicit: yes | epoch: 0.01]

Why it is in scope

Data leakage is a machine learning failure mode where information from outside the training scope — such as test set data, future information, or target variable correlates — inadvertently enters the training process. It produces overoptimistic performance estimates and is prevented through careful data partitioning, proper preprocessing order, and temporal validation protocols. It persists through methodological documentation, validation frameworks, and teaching practices in data science.

Names and aliases

Relations from this entry

  • cmruf04x1013cr671g8zou4lnINSTANCE_OF →

    Leakage (data leakage in statistical/ML practice) is a specific kind of error — specifically, the error of inadvertently including information in training that would not be available at prediction time. A competent speaker in the domain calls leakage 'a type of modeling error.' Nearest kind is error.

Relations to this entry

  • cmsetyrzs05xj3vv3fp6jbzjo← INSTANCE_OF

    Data leakage IS a specific kind of leakage (information escaping unintended channels). In ML, it specifically describes information flow from training to evaluation data — a subtype of the broader leakage construct.

Record identity

Created
Aug 4, 2026, 2:41 PM UTC
Content hash
3f03d3285575029e65de01b89c57d1adb189f8fa8e26e8a8232af2746dfd784c

Open a related act record