Normalization (statistics) - Wikipedia

statisticsdata-preprocessingmachine-learningfeature-scaling

Abstraction: Rescaling data to a common scale for comparison

Key points:

Normalization broadly means adjusting values on different scales to a common notional scale; includes standardization (z-score), min-max scaling, and quantile normalization.
The z-score (standard score) = (value − mean) / std dev; maps any distribution to approximately standard normal when population parameters are known.
William Gosset (publishing as "Student") introduced the t-statistic in 1908 for small-sample normalization; Fisher later formalized it.
Batch normalization was proposed by Ioffe and Szegedy in 2015 to accelerate deep neural network training by reducing internal covariate shift.
Min-max scaling maps values to [0, 1]: x' = (x − min) / (max − min); used widely in machine learning feature preprocessing.
Normalization only makes sense for ratio measurements (where ratios are meaningful), not interval measurements.

Connections: Ronald Fisher · Karl Pearson · Normalization · Standardization · Batch Normalization · Feature Scaling