Method and system for causal modeling and outlier detection

• US 8,140,301 B2
• Filed: 04/30/2007
• Issued: 03/20/2012
• Est. Priority Date: 04/30/2007
• Status: Expired due to Fees

• Current Assignee: GlobalFoundries, Inc.
• Original Assignee: International Business Machines Corporation
• Inventors: Abe, Naoki, Jensen, David L., Merugu, Srujana, Wong, Justin Wai-Chow
• Primary Examiner(s): Rodriguez, Paul
• Assistant Examiner(s): PIERRE LOUIS, ANDRE
• Application Number:
  US11/742,227
• Publication Number:
  US 20080270088A1
• Time in Patent Office: 1,786 Days
• Field of Search: 703/2, 706/15, 706/25
• US Class Current: 703/2
• CPC Class Codes: G05B 17/02 electric

Abstract

A method (and system) for causal modeling includes modeling a data set. The modeling includes estimating a reverse Bayesian forest for the data set and detecting outliers in a separate data set. Detecting the outliers includes applying the reverse Bayesian forest to the separate data set to obtain a probability value assigned to data points in the separate data set and identifying outliers in the separate data set by evaluating the probability value given by the reverse Bayesian forest.

Claims

1. A method, implemented on a computer, for causal modeling, comprising:

modeling, using the computer, a data set, said modeling comprising estimating a reverse Bayesian forest for the data set, said reverse Bayesian forest comprising a set of tree structures, wherein in each of said tree structures a direction of edges leads from a root of the tree structures towards leaves of the tree structures;

detecting outliers in a separate data set, said detecting comprising;

applying the reverse Bayesian forest to the separate data set to obtain a probability value assigned to data points in the separate data set; and

identifying outliers in the separate data set by evaluating the probability value given by the reverse Bayesian forest; and

determining the root cause of the outliers by searching for the outliers in the reverse Bayesian forest and determining root cause feature dependencies for the outliers.

2. The method of claim 1, wherein said identifying outliers comprises determining whether the probability value is lower than a predetermined threshold.

3. The method of claim 1, wherein said tree structures comprise a set of directed edges from a first set of variables to a second variable represents a conditional probability density of the second variable given the first set of variables.

4. The method of claim 3, wherein said conditional probability density is represented by a linear Gaussian model.

5. The method of claim 1, further comprising associating a model of unconditional density for leaf nodes of said reverse Bayesian forest.

6. The method of claim 1, wherein the reverse Bayesian forest comprises a single Bayesian network.

7. The method of claim 1, wherein said estimating the reverse Bayesian forest comprises estimating a structure of the reverse Bayesian forest and probability parameters associated with the Bayesian forest.

8. The method of claim 1, wherein said estimating a reverse Bayesian forest comprises determining a optimal model within a class of reverse Bayesian forests.

9. A computer system for causal modeling, comprising a computer processing apparatus, the computer processing apparatus comprising:

a modeling unit configured to estimate a reverse Bayesian forest for the data set, the term—

using the computer processing apparatus, said reverse Bayesian forest comprising a set of tree structures, wherein in each of said tree structures a direction of edges leads from a root of the tree structures towards leaves of the tree structures;

a detecting unit configured to detect outliers in a separate data set by;

applying the reverse Bayesian forest to the separate data set to obtain a probability value assigned to data points in the separate data set; and

identifying outliers in the separate data set by evaluating the probability value given by the reverse Bayesian forest; and

a determining unit configured to determine the root cause of the outliers by searching for the outliers in the reverse Bayesian forest and determining root cause feature dependencies for the outliers.

10. The computer system of claim 9, wherein said detecting unit identifies outliers by determining whether the probability value assigned is lower than a predetermined threshold.

11. The computer system of claim 9 wherein said tree structures comprise a set of directed edges from a first set of variables to a second variable represents a conditional probability density of the second variable given the first set of variables.

12. The computer system of claim 11, wherein said conditional probability density is represented by a linear Gaussian model.

13. The computer system of claim 9, wherein said detecting unit further associates a model of unconditional density for leaf nodes of said reverse Bayesian forest.

14. The computer system of claim 9, further comprising a storage unit storing training data,

wherein the modeling unit estimates the reverse Bayesian forest based on the training data.

15. A non-transitory computer readable storage device embodying a program, executable by a digital processing system, for causal modeling and outlier detection, said program comprising:

instructions for modeling a data set, said modeling comprising estimating a reverse Bayesian forest for the data set, said reverse Bayesian forest comprising a set of tree structures, wherein in each of said tree structures a direction of edges leads from a root of the tree structures towards leaves of the tree structures;

instructions for detecting outliers in a separate data set, said detecting comprising;

applying the reverse Bayesian forest to the separate data set to obtain a probability value assigned to data points in the separate data set; and

identifying outliers in the separate data set by evaluating the probability value given by the reverse Bayesian forest; and

instructions for determining the root cause of the outliers by searching for the outliers in the reverse Bayesian forest and determining root cause feature dependencies for the outliers.

16. The non-transitory computer readable storage device of claim 15, wherein said identifying outliers comprises determining whether the probability value is lower than a predetermined threshold.