1
$\begingroup$

Im currently working on an anomaly detection while making a transaction. As a part of the data that I extracted, I have the IP addresses of the indivduals who made the transaction. Since the IP Address doesn't have a coherent meaning behind it and it is arbitrary but it indeed acts as a good indicator for user activity. So my question is how can I convert these IP addresses into input which can be used as input for machine learning purposes (which is suitable for data preprocessing such as encoding and PCA). I know tools like OneHot encoding can do the trick however OneHotEncoder can create multi dimensional data and it gets complex if the dataset is huge + standardizing/normalizing it, is painful. So I'm looking for a better way to encode IP addresses.

$\endgroup$

1 Answer 1

0
$\begingroup$

The domain of all possible random IP addresses is astronomically large, and they have no relation to each other. So any given IP address only provides information about that specific IP address, making it a useless feature for the model when trying to generalize to unseen data with different IP addresses.

However, there might be relevant meta-features derived using the IP address, like location or the number and frequency of an individual's past transactions.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.