Anomaly Detection Use Case

[9race]

Hi all,

Thanks for this open source library. I was wondering if there's any way to perform anomaly detection with Chronos?

Thanks!

[abdulfatir]

@9race you can potentially use Chronos for forecasting-based anomaly detection, but we haven't tried it.

[mxsrv]

My team and I are currently trying to find a way to do so. Our idea was either to predict the delta t between 2 anomalies, i.e. transform the dataset into a list of delta t values, or to transform the dataset into a "binary" classification, where all non-anomalies are zero and all anomalies are non-zero. This tends to work pretty okayish when using a dataset with a lot of anomalies and a limited amount of data points overall. We used a synthetic sine curve where we set all values >0.9/>-0.9 as an anomaly. Unfortunately as the maximum recommended prediction length is 64, it is very unlikely that an anomaly is present in there.

Do you have any idea how we could proceed here?