The roots of normalizing and scaling can be better understood by examining the concept of unit vectors in introductory linear algebra. Please see the following links for some common references for unit vectors:
- You can refer to unit vectors at https://en.wikipedia.org/wiki/Unit_vector
- For scalar, you can refer to https://en.wikipedia.org/wiki/Scalar_(mathematics)
In the case of input sensitive algorithms, such as SVM, it is recommended that the algorithm be trained on scaled values (for example, range from 0 to 1) of the features rather than the absolute values as represented by the original vector.