Any case studies about anomaly detection in time series data?

Hi !
I ask if any of you know of any examples in pymc3 about this type of cases (mainly fraud in the financial market…)

Thankyou all :slight_smile:

This is a broad area of research - are there any specific classes of models or data that you are especially interested in? One of the tutorial docs shows how to fit a stochastic volatility model to stock returns; while it may not be immediately apparent how this connects to financial fraud, it suggests a powerful modeling strategy. You can assume that there is some latent process generating the data (potentially the state of the market or the fraudulent/non-fraudulent status of some transactions) and then use the data to determine when the system switches states.

Yes ! I’m interesting about detection of money laundering between a normal and a criminal account. Do you know of any specific case by chance?
I find the link you sent me very interesting, thank you.

I can be most useful if you provide me a clearer idea of what kind of data you are trying to model. Is it integer-valued daily counts of transactions? Positive-valued sizes of transactions? Do you happen to have additional predictors as well? Are there any clearly labeled fraudulent accounts to learn from? Are these data located on a network? Any other clarifications are helpful!