Machine Learning

Data & Splits

Train/validation/test the right way and avoid leakage.

Train / Validation / Test

  • Train: fit model parameters.
  • Validation: tune hyperparameters / select features.
  • Test: final unbiased evaluation (use once at the end).

Leakage Checklist

  • Preprocessing fitted on full dataset (scale/encode before splitting).
  • Using future information to predict the past (time series).
  • Target-derived features (anything computed using y).
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(
  X, y, test_size=0.2, random_state=42, stratify=y
)