Using a custom anomaly detection model

I read in the official blog post that the roadmap for anomaly detection included the ability to include your own model. I wanted to know what’s the current state of this feature. Is it something we can expect in the short term?

Consider the following use-case:
I have an anomaly detection model I created which is custom-made for the business logic of the application. It is not something that the current model notices so I would like to expand upon the existing functionality by plugging in my own model.

Thank you

hi, @amir

Thanks for your interest in AD. We have no plan to release this feature in short term. But we are interested in your use case. Do you already have some custom model to do anomaly detection? If you don’t mind, can you explain more about this model? Want to learn more about how user is building/training a custom model, how to persist the custom model (we can support format your model is using, so you can directly upload your model)? Will your model be trained along with feeding streaming data or you need to train the model offline to adopt new data patterns and re-upload it to apply.

Thanks

Hi @ylwu,

We have created a custom model for anomaly detection in user behavior using RapidMiner software. We have build it using a combination of unsupervised machine learning algorithms in order to detect the patterns of anomalous user behavior in large dataset. Then, we have trained supervised machine learning algorithm to learn the pattern and classify the data on intake. Regarding the deployment, we would export the model from RapidMiner as .pmml and include it in custom Java application. We would re-train this specific model offline and re-upload it.

Thank you

1 Like

hi, @amir

Appreciate your sharing. We will research more about supporting custom model. Will put it on ODFE roadmap when we plan to do it.

1 Like