Outlier

Siddhi enables users to identify outliers using linear regression on real time, data streams. The outlier function takes in a dependent event stream (Y), an independent event stream (X) and a user specified range for outliers, and returns whether the current event is an outlier, based on the regression equation that fits historical data.

Input Parameters

Parameter	Required/Optional	Description
Calculation Interval	Optional	The frequency of regression calculation. Default value: 1 (i.e. at every event)
Batch Size	Optional	The maximum number of events used for a regression calculation Default value: 1,000,000,000 events
Confidence Interval	Optional	Confidence Interval to be used for regression calculation Default value: 0.95
Range	Required	Number of standard deviations from the regression equation
Y Stream	Required	Data stream of the dependent variable
X Stream	Required	Data stream of the independent variable

Output Parameters

Parameter	Name	Description
Outlier	outlier	True if the event is an outlier, False if not
Standard Error	stdError	Standard Error of the Regression Equation
β coefficients	beta0, beta1	β coefficients of the Regression Equation
Input Stream Data	Name given in the input stream	All items sent in the input stream

Examples

The following query submits the number of standard deviations to be used as a range (2), a dependent input stream (Y) and an independent input stream X, that will be used to perform linear regression between Y and X and output whether the current event is an outlier or not.

from StockExchangeStream#transform.timeseries:outlier(2, Y, X)

select *

insert into StockForecaster

When executed, the above query will return whether the current event is an outlier or not along with the standard error of the regression equation (ε), β coefficients and all the items available in the input stream.