Base120 · Decomposition
Dimensional Reduction
Focus on most informative variables while discarding noise or redundancy
When to use
"Use when a dataset has many variables but only a few carry the signal. Reduce dimensions to the most informative variables, discarding noise and redundancy. Improves both understanding and computational efficiency."
Example
"A customer dataset has 200 variables (demographics, behavior, transactions). PCA reveals that 12 principal components capture 95% of the variance. The 12 components are more interpretable and more computationally tractable than the 200 raw variables."
Common misuse
"Discarding dimensions without checking whether they carry signal in interaction. A variable that is noise alone may be signal in combination with another variable. Test interactions before discarding."