AI Data Leakage
Fit preprocessing statistics on training data only.
How it works
Leakage happens when a model or a modelling decision uses information unavailable in its intended evaluation setting. Preprocessing can leak even without target labels: a mean computed from held-out features influences the transformation. Keep fitted preprocessing inside a training pipeline and apply its learned parameters to later records.
Run the code
Try in playground loads this complete standard-library Python example without executing it. Choose Run when ready. Local review uses Python 3.12.14; the existing browser editor uses Pyodide 0.27.7. These small fictional examples need no account, external data, model-provider key or database service.
Example
from statistics import mean
train = [2.0, 4.0, 6.0]
test = [100.0]
training_mean = mean(train)
print("training mean:", training_mean)
print("transformed held-out:", [v-training_mean for v in test])
print("leaking combined mean:", mean(train+test))
Expected result
training mean: 4.0 transformed held-out: [96.0] leaking combined mean: 28.0
The held-out value changes the combined mean from four to twenty-eight. That is precisely why it must not help fit the training transformation. An extreme future value should remain visibly extreme rather than making the training data look more centred.