Yeo-Johnson Transformation

The Yeo-Johnson transformation is a statistical power transformation that can make a numerical feature more symmetric and stabilize its variance. Unlike the related Box-Cox method, it can be applied to positive, zero, and negative values. A parameter controls the transformation shape and is often estimated from training data. Machine learning pipelines may use it before regression, clustering, or methods that benefit from approximately normal and consistently scaled features. The fitted transformation must be applied unchanged to validation and production data to avoid leakage and inconsistency. It does not fix outliers, measurement errors, or every nonlinear relationship, so teams evaluate whether the transformed feature actually improves model behavior and interpretability.