# Data leakage in model evaluation

Source: https://onco.cc/terms/data-leakage/  
OnCo record `data-leakage` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Leakage is when information from the test data reaches the model during training, through a scaler fitted on all samples, a patient split across folds or a benchmark that appeared in the pretraining set, making the model look better than it is.

## Summary

In statistics and machine learning, leakage is the use of information during training that would not be available at prediction time, which produces overly optimistic performance estimates (Wikipedia). Common cancer-data forms are normalising or selecting genes on the full cohort before splitting, duplicated patients across TCGA and cBioPortal studies, and benchmark contamination, where a foundation model's pretraining corpus contained the public test set (TCGA slides or profiles), so that its downstream score is partly recall.

## Fields

- Kind: Term
- Last checked: 2026-09-24
- Also known as: data leakage; leakage; train-test leakage; cross-validation leakage; leaky preprocessing; fit on train fold only; benchmark contamination; test set contamination
- Tags: cansim-terms

## Notes

- Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/cross-validation-leakage.

## Sources

- Wikipedia: https://en.wikipedia.org/wiki/Leakage_(machine_learning)
- Wikipedia: https://en.wikipedia.org/wiki/Leakage_(machine_learning)

## Connected records

- terms: [Batch effects and harmonisation](https://onco.cc/terms/batch-effects/), [Benchmarks, leaderboards and contamination](https://onco.cc/terms/leaderboard-benchmark/), [Cancer AI vocabulary (CanSim terms map)](https://onco.cc/terms/cancer-ai-vocabulary/), [Cross-validation and stratified k-fold](https://onco.cc/terms/cross-validation/), [Drug-response data splits: leave-cell-line-out, leave-drug-out, leave-tissue-out](https://onco.cc/terms/drug-response-splits/), [Principal component analysis (PCA) as a feature compressor](https://onco.cc/terms/pca/), [Train, validation and test split discipline](https://onco.cc/terms/train-test-discipline/)

---
JSON: https://onco.cc/api/v1/entities/data-leakage.json