Method and apparatus for representing and generating evaluation functions in a data classification system

• US 20030037016A1
• Filed: 07/16/2001
• Published: 02/20/2003
• Est. Priority Date: 07/16/2001
• Status: Abandoned Application

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Brodie, Mark, Rish, Irina, Vilalta, Ricardo, Oblinger, Daniel
• Application Number:
  US09/906,168
• Publication Number:
  US 20030037016A1
• Time in Patent Office: Days
• US Class Current: 706/47
• CPC Class Codes: G06F 18/24323 Tree-organised classifiers
• CPC Class Codes: G06N 5/025 Extracting rules from data

Abstract

A unified framework is disclosed for representing and generating evaluation functions for a classification system. The disclosed unified framework provides evaluation functions having characteristics of both traditional or purity-based evaluation functions (class uniformity) and discrimination-based evaluation functions (discrimination power). The disclosed framework is based on a set of configurable parameters and is a function of the distance between examples. By varying the choice of parameters and the distance function, more emphasis is placed on either the class uniformity or the discrimination power of the induced example subsets. A user-configurable function is used to score each of the features based on the class uniformity and discrimination power measures and thereby select the feature having a highest score to partition the data (e.g., using a decision tree or rule-base). This process is recursively applied until all of the examples are partitioned.

Claims

1. A method for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, said method comprising the steps of:

establishing an evaluation function to partition said domain dataset, wherein said evaluation function includes a class uniformity measure and a discrimination power measure, and a weight for each of said class uniformity and discrimination power measures; and

partitioning said domain dataset using said evaluation function.

2. The method of claim 1, further comprising the step of obtaining a model that may be used to classify additional datasets.

3. The method of claim 1, wherein said partitioning step establishes nodes in a decision tree.

4. The method of claim 1, wherein said feature may be a conjunction of features and said partitioning step establishes rules for a rule-based classification system.

5. The method of claim 1, wherein said class uniformity measure is obtained by comparing each example in said domain dataset to other examples in said domain dataset;

and obtaining a first count of a number of examples having a same feature value and same class value.

6. The method of claim 5, further comprising the step of offsetting said first count by a second count of a number of examples having a same feature value and a different class value.

7. The method of claim 1, wherein said discrimination power measure is obtained by comparing each example in said domain dataset to other examples in said domain dataset;

and obtaining a third count of a number of examples having a different feature value and a different class value.

8. The method of claim 7, further comprising the step of offsetting said third count by a fourth count of a number of examples having a different feature value and a same class value.

9. The method of claim 1, wherein said evaluation function further comprises a weight distance, α

, that establishes a relative importance of the distance between any two examples.

10. A method for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, said method comprising the steps of:

evaluating a class uniformity measure for each of said examples for every feature value;

evaluating a discrimination power measure for each of said examples for every feature value;

determining a score for each of said features using a function that considers both said class uniformity measure and said discrimination power measure;

selecting a feature having a highest score to use to partition said data; and

recursively applying said two evaluating steps and said determining and selecting steps until all of said examples are partitioned.

11. The method of claim 10, wherein said selecting step establishes a node in a decision tree.

12. The method of claim 10, wherein said feature may be a conjunction of features and said selecting step establishes a rule for a rule-based classification system.

13. The method of claim 10, wherein said partitioned examples provide a model that may be used to classify data.

14. The method of claim 10, wherein said step of evaluating a class uniformity measure further comprises the step of:

comparing each example in said domain dataset to other examples in said domain dataset; and

obtaining a first count of a number of examples having a same feature value and same class value.

15. The method of claim 14, further comprising the step of offsetting said first count by a second count of a number of examples having a same feature value and a different class value.

16. The method of claim 10, wherein said step of evaluating a discrimination power measure further comprises the step of:

comparing each example in said domain dataset to other examples in said domain dataset; and

obtaining a third count of a number of examples having a different feature value and a different class value.

17. The method of claim 16, further comprising the step of offsetting said third count by a fourth count of a number of examples having a different feature value and a same class value.

18. The method of claim 10, further comprising the step of varying a weight vector, θ

, to establish a weight for each of said class uniformity and discrimination power measures.

19. The method of claim 10, further comprising the step of varying a weight distance, α

, to establish a relative importance of the distance between any two examples.

20. A method for establishing an evaluation function for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, said method comprising the steps of:

providing one or more configurable parameters that evaluate a class uniformity measure and a discrimination power measure and provide a weight for each of said class uniformity and discrimination power measures; and

providing a configurable function that is based on said class uniformity measure and said discrimination power measure to determine a score for each of said features, said score used to identify a feature to partition said domain dataset.

21. The method of claim 20, wherein said class uniformity measure is obtained by comparing each example in said domain dataset to other examples in said domain dataset;

and obtaining a first count of a number of examples having a same feature value and same class value.

22. The method of claim 21, further comprising the step of offsetting said first count by a second count of a number of examples having a same feature value and a different class value.

23. The method of claim 20, wherein said discrimination power measure is obtained by comparing each example in said domain dataset to other examples in said domain dataset;

and obtaining a third count of a number of examples having a different feature value and a different class value.

24. The method of claim 23, further comprising the step of offsetting said third count by a fourth count of a number of examples having a different feature value and a same class value.

25. The method of claim 20, wherein said evaluation function further comprises a weight distance, α

, that establishes a relative importance of the distance between any two examples.

26. A system for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a memory that stores computer-readable code; and

a processor operatively coupled to said memory, said processor configured to implement said computer-readable code, said computer-readable code configured to;

establish an evaluation function to partition said domain dataset, wherein said evaluation function includes a class uniformity measure and a discrimination power measure, and a weight for each of said class uniformity and discrimination power measures; and

partition said domain dataset using said evaluation function.

27. A system for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a memory that stores computer-readable code; and

a processor operatively coupled to said memory, said processor configured to implement said computer-readable code, said computer-readable code configured to;

evaluate a class uniformity measure for each of said examples for every feature value;

evaluate a discrimination power measure for each of said examples for every feature value;

determine a score for each of said features using a function that considers both said class uniformity measure and said discrimination power measure;

select a feature having a highest score to use to partition said data; and

recursively apply said two evaluating steps and said determining and selecting steps until all of said examples are partitioned.

28. A system for establishing an evaluation function for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a memory that stores computer-readable code; and

a processor operatively coupled to said memory, said processor configured to implement said computer-readable code, said computer-readable code configured to;

provide one or more configurable parameters that evaluate a class uniformity measure and a discrimination power measure and provide a weight for each of said class uniformity and discrimination power measures; and

provide a configurable function that is based on said class uniformity measure and said discrimination power measure to determine a score for each of said features, said score used to identify a feature to partition said domain dataset.

29. An article of manufacture for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a computer readable medium having computer readable code means embodied thereon, said computer readable program code means comprising;

a step to establish an evaluation function to partition said domain dataset, wherein said evaluation function includes a class uniformity measure and a discrimination power measure, and a weight for each of said class uniformity and discrimination power measures; and

a step to partition said domain dataset using said evaluation function.

30. An article of manufacture for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a computer readable medium having computer readable code means embodied thereon, said computer readable program code means comprising;

a step to evaluate a class uniformity measure for each of said examples for every feature value;

a step to evaluate a discrimination power measure for each of said examples for every feature value;

a step to determine a score for each of said features using a function that considers both said class uniformity measure and said discrimination power measure;

a step to select a feature having a highest score to use to partition said data; and

a step to recursively apply said two evaluating steps and said determining and selecting steps until all of said examples are partitioned.

31. An article of manufacture for establishing an evaluation function for partitioning a domain dataset, said domain dataset having a plurality of examples, each of said examples characterized by at least one feature and one class value, said feature having a plurality of possible feature values, comprising:

a computer readable medium having computer readable code means embodied thereon, said computer readable program code means comprising;

a step to provide one or more configurable parameters that evaluate a class uniformity measure and a discrimination power measure and provide a weight for each of said class uniformity and discrimination power measures; and

a step to provide a configurable function that is based on said class uniformity measure and said discrimination power measure to determine a score for each of said features, said score used to identify a feature to partition said domain dataset.