Identifying influential effects to be adjusted in goal seek analysis

Patent metadata

Field	Value
Publication number	US-12056622-B2
Application number	US-202117166793-A
Country	US
Kind code	B2
Filing date	Feb 3, 2021
Priority date	Feb 3, 2021
Publication date	Aug 6, 2024
Grant date	Aug 6, 2024

Abstract

A method for identifying influential effects that contribute most to a status change of a target index for goal seeking analysis. The method includes generating a candidate list of significant changed predictors between the normal and abnormal status time periods in collected data, and building a plurality of regression models from the collected data. The method determines a first value (trend value or Pearson correlation value) for each of the significant changed predictors based on whether at least one of the significant changed predictors have a significant change trend using the regression models. The method obtains a second predictor importance value for each of the significant changed predictors from a single model built on all the collected data. The method generates a final predictor value for each of the significant changed predictors by combining the first value with the second predictor importance value for each of the significant changed predictors.

First claim

What is claimed is:

1. A computer-implemented method for identifying influential effects that contribute most to a status change of a target index, the method comprising:
collecting, by at least one processor, data comprising a plurality of predictors corresponding to a plurality of time periods, the collected data indicating a target index value for each time period in the plurality of time periods;
identifying, by the at least one processor, each of the time periods as being one of a normal status or an abnormal status based on the target index value for each of the time periods;
generating, by the at least one processor, a candidate list comprising significant changed predictors between the time periods having the normal status and the time periods having the abnormal status;
building, by the at least one processor, a plurality of regression models from the collected data;
obtaining, by the at least one processor, a first predictor importance value for each of the significant changed predictors from each regression model in the plurality of regression models to produce a set of predictor importance values for each of the significant changed predictors;
generating, by the at least one processor, a series based on the set of predictor importance values for each of the significant changed predictors;
computing, by the at least one processor, a trend value of each of the series of the significant changed predictors;
determining, by the at least one processor, whether at least one of the significant changed predictors have a significant change trend based on the trend value for each of the significant changed predictors;
building, by the at least one processor, a single model from the collected data comprising all the plurality of time periods;
obtaining, by the at least one processor, a second predictor importance value for each of the significant changed predictors from the single model;
combining, by the at least one processor, the first predictor importance value and the second predictor importance value for each of the significant changed predictors to produce a final predictor value for each of the significant changed predictors; and
sorting, by the at least one processor, the final predictor value for each of the significant changed predictors from high to low to identify the significant changed predictors that contribute most to the status change of the target index from the normal status to the abnormal status.

2. The computer-implemented method of claim 1 , wherein the first predictor importance value is the trend value in response to determining that at least one of the significant changed predictors have a significant change trend.

3. The computer-implemented method of claim 1 , wherein generating the candidate list comprises:
comparing, by the at least one processor, a value distribution for each of the plurality of predictors between the normal status and the abnormal status using a nonparametric test;
comparing, by the at least one processor, a series trait for each of the plurality of predictors between the normal status and the abnormal status; and
identifying, by the at least one processor, a predictor as a significant changed predictor for the candidate list when at least one of the value distribution or the series trait for the predictor between the normal status and the abnormal status is not similar based on a predetermined threshold.

4. The computer-implemented method of claim 1 , wherein each regression model in the plurality of regression models comprises a same internal time interval, the plurality of regression models built from the time periods having the normal status to the time periods having the abnormal status.

5. The computer-implemented method of claim 1 , wherein the first predictor importance value is a Pearson correlation value in response to determining that none of the significant changed predictors have a significant change trend.

6. The computer-implemented method of claim 5 , wherein the computer-implemented method determines the Pearson correlation value by selecting a first regression model having the normal status from the plurality of regression models, and computing the Pearson correlation value for each of the significant changed predictors in the candidate list in the first regression model.

7. The computer-implemented method of claim 6 , wherein computing the Pearson correlation value for each of the significant changed predictors in the candidate list in the first regression model comprises for each of the significant changed predictors in the candidate list, selecting a significant changed predictor in the candidate list, and replacing a set of values of the significant changed predictor in the first regression model with a second set of values from one of the time periods having the abnormal status, while maintaining all other field values unchanged.

8. A system configured to identify influential effects that contribute most to a status change of a target index, the system comprising memory for storing instructions, and a processor configured to execute the instructions to:
collect data comprising a plurality of predictors corresponding to a plurality of time periods, the collected data indicating a target index value for each time period in the plurality of time periods;
identify each of the time periods as being one of a normal status or an abnormal status based on the target index value for each of the time periods;
generate a candidate list comprising significant changed predictors between the time periods having the normal status and the time periods having the abnormal status;
build a plurality of regression models from the collected data;
obtain a first predictor importance value for each of the significant changed predictors from each regression model in the plurality of regression models to produce a set of predictor importance values for each of the significant changed predictors;
generate a series based on the set of predictor importance values for each of the significant changed predictors;
compute a trend value of each of the series of the significant changed predictors;
determine whether at least one of the significant changed predictors have a significant change trend based on the trend value for each of the significant changed predictors;
build a single model from the collected data comprising all the plurality of time periods;
obtain a second predictor importance value for each of the significant changed predictors from the single model;
combine the first predictor importance value and the second predictor importance value for each of the significant changed predictors to produce a final predictor value for each of the significant changed predictors; and
sort the final predictor value for each of the significant changed predictors from high to low to identify the significant changed predictors that contribute most to the status change of the target index from the normal status to the abnormal status.

9. The system of claim 8 , wherein the first predictor importance value is the trend value in response to determining that at least one of the significant changed predictors have a significant change trend.

10. The system of claim 8 , wherein the instructions to generate the candidate list comprises instructions to:
compare a value distribution for each of the plurality of predictors between the normal status and the abnormal status using a nonparametric test;
compare a series trait for each of the plurality of predictors between the normal status and the abnormal status; and
identify a predictor as a significant changed predictor for the candidate list when at least one of the value distribution or the series trait for the p

Assignees

• IBM

Inventors

• Xu Jing James
• Gao Lei
• Zhang A Peng
• Wang Rui
• Han Si Er
• Ma Xiao Ming

Classifications

• G06N5/04 Primary

  Inference or reasoning models · CPC title

• G06N20/00 Primary

  Machine learning · CPC title