What is data normalization bias in large-scale data mining pipelines?

Normalization bias occurs when scaling methods distort feature distributions and relationships.

In large-scale data mining, normalization methods like min-max scaling or z-score normalization assume stable distributions. However, when data is skewed, heavy-tailed, or streaming, normalization can distort meaningful variance, compress rare but important signals, and artificially inflate common patterns. This leads to biased downstream models, especially in clustering and anomaly detection. The issue becomes more severe when normalization is applied globally instead of per-segment or per-time-window.