Back to results
US 11,768,915 B1Granted
System, method, and computer program for anomaly detection in time-series data with mixed seasonality
AssigneeAmdocs Development Limited
InventorsGilli Shama, Yaniv Avrashi, Nadeem Azaizah, Pavel Vodolazov
Life of this patent
Estimating the term…
Abstract
As described herein, a system, method, and computer program are provided for anomaly detection in time-series data with mixed seasonality. In use, time-series data of a mixed seasonality type is received. Additionally, the time-series data is segmented, by a defined unit of time, to form a plurality of time-series data segments. Further, the plurality of time-series data segments are processed to determine one or more patterns across the plurality of time-series data segments. Still yet, the one or more patterns are stored and used to perform pattern matching for an input time-series.
Claims17
A non-transitory computer readable medium storing computer code executable by a processor to perform a method comprising:
- receiving time-series data of a single time-series having a sequence of data points in a successive time-based order, the time-series data being of a mixed seasonality type by having portions that repeat on a time-wise basis;
- segmenting the time-series data, by a single defined unit of time, to form a plurality of time-series data segments each having a different subset of the data points in the single time-series;
- processing the plurality of time-series data segments to determine one or more patterns across the plurality of time-series data segments;
- storing the one or more patterns with a characterization that includes:
- a standard representative value for each sampled time segment point,
- a lower bound of data in the pattern for a defined confidence interval, and
- an upper bound of data in the pattern for the defined confidence interval,
- wherein the lower bound and the upper bond defined a normal range of data points in the pattern;
- and
- detecting one or more anomalies in an input time-series, based on the stored one or more patterns, by:
- determining a plurality of segments for the input time-series, including for each time point in the input time-series determining a segment in the input time-series from time zero to the time point, and
- for each segment of the plurality of segments:
- (a) determining a distance between the segment in the input time-series and each pattern of the stored one or more patterns,
- (b) selecting a particular pattern of the stored one or more patterns having a least distance to the segment,
- (c) predicting a range of upcoming normal values for the input time-series, wherein the range is predicted as characterized in the particular pattern, and
- (d) detecting an anomaly for the segment based on a data point within the input-time series being out of the range.
13 more elements
A method, comprising:
- receiving time-series data of a single time-series having a sequence of data points in a successive time-based order, the time-series data being of a mixed seasonality type by having portions that repeat on a time-wise basis;
- segmenting the time-series data, by a single defined unit of time, to form a plurality of time-series data segments each having a different subset of the data points in the single time-series;
- processing the plurality of time-series data segments to determine one or more patterns across the plurality of time-series data segments;
- storing the one or more patterns with a characterization that includes:
- a standard representative value for each sampled time segment point,
- a lower bound of data in the pattern for a defined confidence interval, and
- an upper bound of data in the pattern for the defined confidence interval,
- wherein the lower bound and the upper bond defined a normal range of data points in the pattern;
- and
- detecting one or more anomalies in an input time-series, based on the stored one or more patterns, by:
- determining a plurality of segments for the input time-series, including for each time point in the input time-series determining a segment in the input time-series from time zero to the time point, and
- for each segment of the plurality of segments:
- (a) determining a distance between the segment in the input time-series and each pattern of the stored one or more patterns,
- (b) selecting a particular pattern of the stored one or more patterns having a least distance to the segment,
- (c) predicting a range of upcoming normal values for the input time-series, wherein the range is predicted as characterized in the particular pattern, and
- (d) detecting an anomaly for the segment based on a data point within the input-time series being out of the range.
13 more elements
A system, comprising:
- a non-transitory memory storing instructions;
- and
- one or more processors in communication with the non-transitory memory that execute the instructions to perform a method comprising:
- receiving time-series data of a single time-series having a sequence of data points in a successive time-based order, the time-series data being of a mixed seasonality type by having portions that repeat on a time-wise basis;
- segmenting the time-series data, by a single defined unit of time, to form a plurality of time-series data segments each having a different subset of the data points in the single time-series;
- processing the plurality of time-series data segments to determine one or more patterns across the plurality of time-series data segments;
- storing the one or more patterns with a characterization that includes:
- a standard representative value for each sampled time segment point,
- a lower bound of data in the pattern for a defined confidence interval, and
- an upper bound of data in the pattern for the defined confidence interval,
- wherein the lower bound and the upper bond defined a normal range of data points in the pattern;
- and
- detecting one or more anomalies in an input time-series, based on the stored one or more patterns, by:
- determining a plurality of segments for the input time-series, including for each time point in the input time-series determining a segment in the input time-series from time zero to the time point, and
- for each segment of the plurality of segments:
- (a) determining a distance between the segment in the input time-series and each pattern of the stored one or more patterns,
- (b) selecting a particular pattern of the stored one or more patterns having a least distance to the segment,
- (c) predicting a range of upcoming normal values for the input time-series, wherein the range is predicted as characterized in the particular pattern, and
- (d) detecting an anomaly for the segment based on a data point within the input-time series being out of the range.
16 more elements
Dependencies are inferred from the claim text.