Normalization of input features on anomaly detection

Hi all,

I’d like to know if performing normalization of the input features could have a positive or negative impact on anomaly detection algorithms or no impact at all.

Please share your thougths.

Thank you.

Hi @pguerrero,

With the anomaly detection algorithm presented in the lecture, pre-normalizing the features won’t impact. In the other words, whether or not there is any benefit depends on your choice of algorithm. For example, as explained in Course 1 Week 2 (refer to video feature scaling part 1 & part 2), a neural network based algorithm needs normalization.

Cheers,
Raymond

In this course, the primary use for normalization is to speed up the solution process (such as gradient descent), by adjusting the features so they all have similar magnitudes.

Since anomaly detection doesn’t use gradient descent, there’s no benefit.