I'm currently working on LSTM autoencoder for anomaly detection. My main problem is I have multiple time series - each individual time series corresponds to a different customer, detailing their sales data from 2020 to 2023. I don't want to aggregate data (I want detect anomaly for each customer) neither make model per customer (thousands of customers means thousands of models)

Example of data:

date customer sales
05.02.2020 cust1 15
05.02.2020 cust2 250
05.02.2020 cust3 1
06.02.2020 cust1 26
06.02.2020 cust2 120
06.02.2020 cust3 0

(obviously there are much more customers than in this simple example)

Please guide me on how to incorporate customer-specific information into the LSTM autoencoder model. How I can model my LSTM autoencoder to add layer about type of customer? Or how to model the data so the model will know that each sequence is from different customer? Or maybe for this kind of data is much more suitable model than LSTM autoencoder?

Thank you in advance for your help!

  • $\begingroup$ What kind of anomalies are you looking for? $\endgroup$
    – Jon Nordby
    Oct 17, 2023 at 11:06
  • $\begingroup$ @JonNordby any anomalies in sales, my data is not labeled, so any maybe outliar $\endgroup$
    – suziex
    Oct 24, 2023 at 14:32


Your Answer

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