normalization

Normalization is a broad term that describes methods for transforming data, measurements, or structures to a common scale or reference form. It appears in statistics, data management, computer science, and the sciences, with the aim of enabling comparison, improving numerical behavior, or reducing redundancy.

In statistics and data processing, common normalization techniques include min-max normalization, which rescales values to a fixed range such as 0 to 1, and z-score standardization, which centers data at zero with unit variance. Other methods include robust scaling and unit-length normalization. The choice depends on data distribution, downstream methods, and sensitivity to outliers. Normalization does not create new information but can affect interpretability and model performance.

In database design, normalization refers to organizing data to minimize redundancy and dependency. It uses normal forms (1NF, 2NF, 3NF, BCNF, and higher). The process typically involves decomposing tables into smaller, related tables according to functional dependencies. Normalization improves data integrity and update efficiency but can increase query complexity and reduce performance, leading to deliberate denormalization in some systems.

In machine learning and signal processing, normalization or scaling of features ensures that different attributes contribute proportionally to models. Techniques include L2 or L1 normalization, min-max scaling, and batch normalization in neural networks, which stabilizes training by normalizing activations within layers.

Normalization also appears in measurement science, where data are adjusted for systematic effects, and in bioinformatics, where gene expression data are normalized to account for sequencing depth. Across domains, proper normalization requires awareness of assumptions and potential biases.