DETERMINING A SEASONAL EFFECT IN TEMPORAL DATA

• US 20120016886A1
• Filed: 07/14/2009
• Published: 01/19/2012
• Est. Priority Date: 07/14/2009
• Status: Active Grant

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Cohen, Ira, Eisenberg, Kobi, Ashkenazi, Liat, Krepel, Avishai, Lyan, Marina, Rozman, Michael
• Granted Patent: US 8,468,161 B2
• Time in Patent Office: Days
• US Class Current: 707/748
• CPC Class Codes: G06F 16/2477 Temporal data queries
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202

Abstract

To determine a seasonal effect in temporal data, for each of a plurality of candidate seasons, the following is performed. An interval is defined for the corresponding candidate season. The interval is divided (108) into plural buckets containing respective sets of the temporal data, and statistical measures for corresponding buckets are computed. The computed statistical measures are used to compute (112) a score for the corresponding candidate season. Scores computed for corresponding candidate seasons are used for identifying which of the candidate seasons represents the seasonal effect of the temporal data.

Claims

1. A method of determining a seasonal effect in temporal data, comprising:

receiving (104) the temporal data relating to an aspect of an enterprise;

for each of a plurality of candidate seasons, performing, by a computer (600);

defining an interval for the corresponding candidate season;

dividing (108) the interval into plural buckets containing respective sets of the temporal data,

computing statistical measures for corresponding buckets, and

using the computed statistical measures to compute (112) a score for the corresponding candidate season,

wherein scores computed for corresponding candidate seasons are used for identifying which of the candidate seasons represents the seasonal effect of the temporal data.

2. The method of claim 1, wherein computing the statistical measures comprises computing statistical measures selected from the group consisting of a median, average, standard deviation, and a probability distribution.

3. The method of claim 1, wherein computing the score comprises computing an error score.

4. The method of claim 3, wherein computing the error score comprises:

within each of the buckets, applying a mathematical function based on temporal data values in the bucket and the statistical measure for the bucket to produce an output value; and

aggregating the output values produced for the buckets to compute the error score.

5. The method of claim 4, wherein applying the mathematical function comprises computing absolute deviations of the temporal data values in the bucket from the statistical measure of the bucket, and summing the absolute deviations for corresponding buckets to produce the corresponding output value.

6. The method of claim 1, wherein computing the statistical measures comprises computing probability mapping functions.

7. The method of claim 6, further comprising computing an entropy based on the probability mapping function for each bucket.

8. The method of claim 7, wherein computing the score is based on the entropies of the buckets.

9. The method of claim 1, further comprising selectively using different techniques for computing the score based on whether the temporal data is continuous or discrete.

10. The method of claim 1, further comprising:

after identifying the season associated with the temporal data, estimating a baseline for the temporal data, wherein estimating the baseline includes;

dividing an interval of the identified season into buckets;

mapping the temporal data into the buckets of the interval of the identified season;

computing statistics based on temporal data values in the buckets of the interval of the identified season; and

computing thresholds based on the statistics.

11. The method of claim 10, further comprising:

receiving a new data sample; and

comparing the new data sample to the thresholds to determine whether the new data sample is normal or abnormal.

12. A computer (600) comprising:

a storage media (608) to store temporal data; and

a processor (604) to;

define a set of candidate seasons of arbitrary length;

for each of the candidate seasons in the set, compute a score based on division of the temporal data into plural buckets that are within an interval of the corresponding candidate season, wherein statistical measures are computed for the corresponding plural buckets to enable computation of the score; and

compare the scores of the corresponding candidate seasons to identify one of the candidate seasons as representative of a seasonality of the temporal data.

13. The computer of claim 12, wherein the processor is configured to further:

for each of the buckets, compute an absolute deviation between each temporal data value in the bucket and the statistical measure of the bucket,

wherein the score is computed based on the absolute deviations of the buckets.

14. The computer of claim 12, wherein the processor is configured to further:

for each of the buckets, compute an entropy based on the statistical measure of the bucket,

wherein the score is computed based on the entropies of the buckets.

15. The computer of claim 12, wherein the temporal data comprises one of continuous data and discrete data acquired without regular sampling intervals.

16. An article comprising at least one computer readable storage medium containing instructions that upon execution cause a computer (600) to:

receive temporal data relating to an aspect of an enterprise;

for each of a plurality of candidate seasons;

define an interval for the corresponding candidate season;

divide the interval into plural buckets containing respective sets of the temporal data,

compute statistical measures for corresponding buckets, and

use the computed statistical measures to compute a score for the corresponding candidate season,

wherein scores computed for corresponding candidate seasons are used for identifying which of the candidate seasons represents a seasonal effect of the temporal data.

17. The article of claim 16, wherein the instructions upon execution cause the computer to further:

compute absolute deviations based on temporal data values in each bucket and the corresponding statistical measure of the bucket,

wherein the score is computed based on the absolute deviations.

18. The article of claim 16, wherein the instructions upon execution cause the computer to further:

compute an entropy based on temporal data values in each bucket and the corresponding statistical measure of the bucket,

wherein the score is computed based on the entropies.

19. The article of claim 18, wherein the statistic measure is a probability mass function.

20. The article of claim 16, wherein the instructions upon execution cause the computer to further selectively use different techniques for computing the score based on whether the temporal data is continuous or discrete.