Normaalointi

Normaalointi, or normalization, is a process used across disciplines to adjust data to a common scale or structure. Its goal is to reduce biases caused by differing ranges or formats, enabling fair comparisons, improved numerical stability, and more effective computation.

In statistics and data processing, normalization typically refers to rescaling numeric features. Common methods include min–max normalization, which maps values to a fixed range such as [0, 1]; z-score standardization, which centers data at zero with unit variance; and unit-length (L2) normalization, which divides vectors by their length. Robust scaling uses percentiles to minimize outlier effects.

In machine learning practice, the choice of normalization method depends on the algorithm and data. Some models assume standardized inputs; others benefit from bounded ranges. It is standard to fit normalization parameters on training data only, then apply to validation and test data, to avoid data leakage.

In database design, normalisointi (database normalization) is a separate concept focused on structuring tables to reduce redundancy and update anomalies. Typical goals are to separate related data into related tables and to enforce data integrity through normal forms such as 1NF, 2NF, 3NF and BCNF.

Normalization also appears in signal processing, where amplitudes are scaled to a standard level to avoid clipping or to compare signals, and in other fields where consistent units or formats are required. Overall, normaalointi denotes the broader practice of making data or structures comparable and stable by bringing them to a common standard.