← Back to research

When to Use Standardization and Normalization: Empirical Evidence From ML Models

Scale or standardize?
Data transformed with purpose—
Models learn better.
Machine learning
Data preprocessing
Standardization
Normalization

K. M. Sujon, R. B. Hassan, Z. T. Towshi, M. A. Othman, M. A. Samad, K. Choi, “When to Use Standardization and Normalization: Empirical Evidence From ML Models,” IEEE Access 12:135300-135314 (2024), doi: 10.1109/ACCESS.2024.3462434

Authors

K. M. Sujon

R. B. Hassan

Z. T. Towshi

M. A. Othman

Md Abdus Samad

K. Choi

Published

September 2024

Doi

10.1109/ACCESS.2024.3462434

Abstract

Data preprocessing is a critical step in machine learning pipelines, yet the choice between standardization and normalization often remains arbitrary. This study provides comprehensive empirical evidence on when to use each technique across various machine learning models and datasets. Our experiments reveal that the optimal preprocessing choice depends on the algorithm type, data distribution, and specific application requirements, offering practical guidelines for practitioners.

Citation

 Add to Zotero 

@article{SujonEtAl:2024,
  Author  = {Sujon, K. M. and Hassan, R. B. and Towshi, Z. T. and Othman, M. A. and Samad, M. A. and Choi, K.},
  Title   = {When to Use Standardization and Normalization: Empirical Evidence From ML Models},
  Journal = {IEEE Access},
  Volume  = {12},
  Pages   = {135300-135314},
  Year    = {2024},
  Doi     = {10.1109/ACCESS.2024.3462434}
}