See Batch Normalization. The measures that serve as input to the classifiers are ideally dimensionless dues to normalization.

References

http://timvieira.github.io/blog/post/2016/05/27/dimensional-analysis-of-gradient-ascent

https://en.wikipedia.org/wiki/Mahalanobis_distance

https://en.wikipedia.org/wiki/Nondimensionalization

https://en.wikipedia.org/wiki/Normalization_(statistics)