Missing Data detection

Our engine will create missing data alerts that are specific to each time series EYER is monitoring. To determine when EYER should alert if it stops receiving data, it takes into account frequency and typical breaks is the stream of data received.

We analyze up to 28 days of historical data to understand what “normal” missing data looks like. These “normal” missing data are the data interruptions that are recurring and therefore should not be alerted.

Our algorithm takes the day of the week into account. This means the maximum expected duration of missing data can vary by day. For example:

• A gap that triggers an alert on a Tuesday might not trigger an alert on a Sunday, if longer gaps are typical on Sundays.

• If a metric has the same reporting frequency every day, the missing-data timeout will stay roughly the same throughout the week.

If an unusually long gap occurred only once in the past 28 days, it may still be treated as an outlier. In that case, it increases the allowed waiting time for that specific weekday.

Currently, we allow up to one outlier per weekday.