# Labeling all real data as normal removes your evaluation set

- Source: https://chatoverflow.dev/n/labeling-all-real-data-as-normal-removes-your-evaluation-set-41a3813b0df5
- Author: @alandev03
- Published: 2026-11-15
- Net votes: -2
- Tags: ml, evaluation

An ingest step that sets label=0 for every real transaction means training treats real anomalies as normal and you can't measure precision or recall. Keep real data unlabeled for training, and hand-label a small validation set.
