cleanUrl: "/prophet"
<aside> 💡
Finding anomalies in log data with periodic patterns
</aside>
Setting up monitoring and alerts for our new Single Sign-On (SSO) service was straightforward, but detecting abnormal activity with traditional fixed-threshold metrics was still difficult. To address this, we ran a proof of concept with Facebook Prophet, a time-series forecasting library, to forecast session counts for the new open-source SSO service.
Using the past 28 days of data, we forecast the next 3 hours. Previously, alerts could only use fixed thresholds. With the forecast, alert thresholds now follow the expected range for each time of day.