Data Forecasting

  • Axibase Time Series Database includes built-in forecasting algorithms that can predict abnormalities based on historical data.
  • The accuracy of predictions and the percentage of false positives/negatives depends on the frequency of data collection, the retention interval, and algorithms.
  • Built-in auto-regressive time series extrapolation algorithms (Holt-Winters, ARIMA, etc.) in ATSD can predict failures at early stages.
  • Dynamic predictions eliminate the need to set manual thresholds.

Forecasting Example with Abnormal Deviation:

Forecast Settings:

  • Forecasting settings can be left in automatic mode for easy setup.
  • ATSD selects the most accurate forecasting algorithm for each time-series separately based on a ranking system.
  • The winning algorithm is used to compute forecasts for the next day, week, or month.
  • Pre-computed forecasts can be used in the rule engine.
  • Basic automatic adhoc forecasting can be used directly in graphs and widgets, and forecasts will be calculated for 1 week with automatic settings.

Key Advanced Settings:

SettingDescription
EnabledThe forecast is enabled, new forecasts will be calculated automatically every 24 hours
MetricThe metric to be forecast, for example, cpu_busy
EntityThe entity that will be used. Exclusive with Entity Group.
Entity GroupThe entity group selected from a drop down. Forecast will be calculated for all entities contained within the Entity Group. Exclusive with Entity Group.
Selection IntervalAmount of historical data analyzed when calculating the forecast.
Full ScanFull scan of historical data, used when there is no fresh data for the past 24 hours from the current entity and metric.
If Full Scan is not set, then ATSD will automatically look for metric keys from the past 24 hours.
Set to true when forecasting historical data that is no longer collected or when generating a forecast using “End Time”.
Averaging IntervalInterval over which the data is normalized. When selecting the ARIMA algorithm, the averaging interval cannot be set to less than 1 hour.
Retention IntervalHow long for the forecast is stored. Forecasts older than the retention interval will be deleted.
Store IntervalForecast time-span. How far into the future the data is forecast.
Auto ParametersATSD uses automatic settings to select the best forecast.
Auto AveragingAutomatic averaging interval determined by ATSD.
AlgorithmHolt-Winters or Arima algorithms.
End TimeUsed to calculate the forecast from an exact point in time. Useful when calculating a forecast for data that is not frequently updated. Possible values described on the End Time page.
Important to select “Full Scan” when forecasting historical data that is no longer collected, if “Full Scan” is not set, then ATSD will automatically look for metric keys from the past 24 hours.
NameUnique forecast identifier.