Calculation of p(x) in Anomaly Detection

Hi all!

I have some doubts I’d like to clarify for the calculation of p(x). Assuming we only have one feature x, p(x) would be calculated using the normal cumulative distribution function (CDF).

Wouldn’t this imply that anomalies in the right tail (right end of the gaussian curve) result in a high p(x), and therefore not be detected?

Edit: I’ve come to realise that it’s the PDF rather than the CDF, hence anomalies at the right tail will result in a low p(x).

So I think you have answered your own question! Nothing’s better than that!


1 Like