I just started my first project where I'm trying to find anomalies in the energy usage of a air conditioner. The only usable data I could obtain was the energy data for a few months. Since the energy usage was mostly discontinuous, I decided that it would be a good idea to compare the cumulative energy usage every day to get a good model.

I read that it would be relatively simpler if we could reduce contextual anomalies to point anomalies. Hence, I interpolated the available data to 30 second intervals so that I can compare the data for every time stamp over the period of months. However, what I did not think about was the fact that I would have to set a probability threshold below which the data point is an anomaly for every 30 seconds (2880 individual threshold values). Is there a way to automatically set the thresholds for all these time stamps? If not, what would be a good way to restart this given the data I have is very limited?

I also realise that it is not possible to get enough labelled data to calculate the F1 scores at every time stamp and gauge the optimal epsilon (threshold) using this.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.