C3_W1_Anomaly Detection_Feature_Distribution

[Scofield_Wu]

I have a general question about the feature distribution used in Anomaly Detection. In the course, we used the Normal distribution PDF to model p(x), if the feature is not Normal, we transform it into a Normal distribution.

However, it is only valid for continuous features, how about categorical features? Are we not allowed to use categorical features to identify/model anomaly behaviours?

[TMosh]

There are statistical methods for categorical data.
Example:

Pearson's chi-squared test