-
What is your workflow for auditing potential data leakage inside your experi...
Method Help
Data leakage is a pervasive but often silent problem. Information from test set accidentally leaks into training procedure and produces over‑optimistic evaluation results. What step‑by‑step audit routine helps you hunt for leakage risks?
Feng Song, Feng Shi and 32 others5 Comments-
Trace every data transformation step and verify no test‑set information is used during training phase.
2 -
Inspect preprocessing scripts: scaling, normalisation must be fit strictly on training split only.
2 -
Test with simple baseline model; suspiciously near‑perfect performance is a red flag.
2 -
Have a second lab member review pipeline code specifically looking for leakage paths.
2 -
Document your splitting strategy clearly in lab notes and manuscript.
2
-