Azure Anomaly Detector - only detects spikes

Question

I am testing anomaly detector on metrics of count of specific event per hour for last 90 days. For some reason I always get spikes (isPositive) only, but never drops, while I'm mostly interested to detect drops. Data has weekly seasonality (expected drops on weekends) and definitely has abnormal drops mid week unusual for this day of week. I also tried to play with specific hours to take them to extremely low for this time and week day. I tried different values for sensitivity (between 90 and 20). On the positive side I get too many spikes, which create a lot of noise, while low sensitivity value didn't help to get rid of them. Below is a link to request JSON. Request JSON

Answer 1

You can try to set the maxRatio to 0.01, it should be what you expect. Currently the sensitivity control is not good enough for low value. But we will rollout a new version in next week to improve it. And you can also leverage https://aka.ms/addemo, and use a CSV to have more test.