Method, program and apparatus for generating two-class classification/prediction model

• US 20090222390A1
• Filed: 05/04/2009
• Published: 09/03/2009
• Est. Priority Date: 11/13/2006
• Status: Active Grant

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Yuta, Kohtarou
• Granted Patent: US 7,725,413 B2
• Time in Patent Office: Days
• US Class Current: 706/13
• CPC Class Codes: G06F 16/244 Grouping and aggregation
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G16C 20/70 Machine learning, data mining or chemometrics

Abstract

A method includes: a) preparing as training data a sample set that contains a plurality of samples belonging to a first class and a plurality of samples belonging to a second class; b) generating, by performing discriminant analysis on the sample set, a first discriminant function having a high classification characteristic for the first class and a second discriminant function having a high classification characteristic for the second class; c) by classifying the sample set using the first and second discriminant functions, isolating any sample whose classification results by the first and second discriminant functions do not match; d) forming a new sample set by grouping together any sample thus isolated, and repeating b) and c) by using the new sample set; and e) causing d) to stop when the number of samples each of whose classification results do not match in c) has decreased to or below a predetermined value.

Claims

1. A program for generating a two-class classification/prediction model, said program causing a computer to perform a process comprising:

a) preparing as training data a sample set that contains a plurality of samples belonging to a first class and a plurality of samples belonging to a second class;

b) generating, by performing discriminant analysis on said sample set, a first discriminant function having a high classification characteristic for said first class and a second discriminant function having a high classification characteristic for said second class;

c) by classifying said sample set using said first and second discriminant functions, isolating any sample whose classification results by said first and second discriminant functions do not match;

d) repeating said b) and c) by using a new sample set which is formed by grouping together any sample isolated in said c); and

e) causing said d) to stop when the number of samples each of whose classification results do not match in said c) has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value.

2. The program according to claim 1, wherein

said first discriminant function is generated by carrying out:

f) generating an initial discriminant function by performing discriminant analysis on said sample set;

g) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any sample misclassified by said initial discriminant function as being a sample belonging to said first class when said sample is actually a sample belonging to said second class; and

h) repeating said g) by using said new discriminant function from said g) as said initial discriminant function, until the number of samples misclassified into said first class by said initial discriminant function decreases to substantially zero, and wherein

said second discriminant function is generated by carrying out;

i) generating an initial discriminant function by performing discriminant analysis on said sample set;

j) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any sample misclassified by said initial discriminant function as being a sample belonging to said second class when said sample is actually a sample belonging to said first class; and

k) repeating said j) by using said new discriminant function from said j) as said initial discriminant function, until the number of samples misclassified into said second class by said initial discriminant function decreases to substantially zero.

3. The program according to claim 1, wherein said e) allows said d) to continue until the number of samples each of whose classification results do not match decreases to substantially zero.

4. The program according to claim 2, wherein said initial discriminant function and said new discriminant function are each generated by performing feature extraction on an initial parameter set pregenerated for said sample set prepared as said training data, thereby forming a final parameter set, and by performing said discriminant analysis using said final parameter set.

5. The program according to claim 1, wherein said first and second discriminant functions determined in said b) after completion of said e) are set up as a classification/prediction model for samples for which it is not known to which class, said first class or said second class, each belongs.

6. A method for generating a two-class classification/prediction model, comprising:

a) preparing as training data a sample set that contains a plurality of samples belonging to a first class and a plurality of samples belonging to a second class;

b) generating, by performing discriminant analysis on said sample set, a first discriminant function having a high classification characteristic for said first class and a second discriminant function having a high classification characteristic for said second class;

c) by classifying said sample set using said first and second discriminant functions, isolating any sample whose classification results by said first and second discriminant functions do not match;

d) repeating said b) and c) by using a new sample set which is formed by grouping together any sample isolated in said c); and

e) causing said d) to stop when the number of samples each of whose classification results do not match in said c) has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value, and wherein

said first and second discriminant functions determined in said b) are set up as a classification/prediction model for samples of unknown classes.

7. The method according to claim 6, wherein said high classification characteristic in said b) provides a classification rate of 95% to 100%.

8. The method according to claim 6, wherein

said first discriminant function is generated by carrying out:

f) generating an initial discriminant function by performing discriminant analysis on said sample set;

g) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any sample misclassified by said initial discriminant function as being a sample belonging to said first class when said sample is actually a sample belonging to said second class; and

h) repeating said g) by using said new discriminant function from said g) as said initial discriminant function, until the number of samples misclassified into said first class by said initial discriminant function decreases to substantially zero, and wherein

said second discriminant function is generated by carrying out;

i) generating an initial discriminant function by performing discriminant analysis on said sample set;

j) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any sample misclassified by said initial discriminant function as being a sample belonging to said second class when said sample is actually a sample belonging to said first class; and

k) repeating said j) by using said new discriminant function from said j) as said initial discriminant function, until the number of samples misclassified into said second class by said initial discriminant function decreases to substantially zero.

9. The method according to claim 6, wherein said e) allows said d) to continue until the number of samples each of whose classification results do not match decreases to substantially zero.

10. The method according to claim 8, wherein said initial discriminant function and said new discriminant function are each generated by performing feature extraction on an initial parameter set pregenerated for said sample set prepared as said training data, thereby forming a final parameter set, and by performing said discriminant analysis using said final parameter set.

11. A method for generating a chemical toxicity prediction model, comprising:

a) preparing as training data a sample set that contains a plurality of chemicals belonging to a first class and a plurality of chemicals belonging to a second class, wherein said chemicals in said first class have a specific kind of toxicity and said chemicals in said second class do not have said toxicity;

b) generating, by performing discriminant analysis on said sample set, a first discriminant function having a high classification characteristic for said first class and a second discriminant function having a high classification characteristic for said second class;

c) by classifying said sample set using said first and second discriminant functions, isolating any chemical whose classification results by said first and second discriminant functions do not match;

d) repeating said b) and c) by using a new sample set which is formed by grouping together any chemical isolated in said c); and

e) causing said d) to stop when the number of chemicals each of whose classification results do not match in said c) has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value, and wherein

said first and second discriminant functions determined in said b) after completion of said e) are set up as a classification/prediction model for chemicals of unknown classes.

12. The method according to claim 11, wherein

said first discriminant function is generated by carrying out:

f) generating an initial discriminant function by performing discriminant analysis on said sample set;

g) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any chemical misclassified by said initial discriminant function as being a chemical belonging to said first class when said chemical is actually a chemical belonging to said second class; and

h) repeating said g) by using said new discriminant function from said g) as said initial discriminant function, until the number of samples misclassified into said first class by said initial discriminant function decreases to substantially zero, and wherein

said second discriminant function is generated by carrying out;

i) generating an initial discriminant function by performing discriminant analysis on said sample set;

j) generating a new discriminant function by performing discriminant analysis on a new sample set which is formed by removing from said sample set any chemical misclassified by said initial discriminant function as being a chemical belonging to said second class when said chemical is actually a chemical belonging to said first class; and

k) repeating said j) by using said new discriminant function from said j) as said initial discriminant function, until the number of samples misclassified into said second class by said initial discriminant function decreases to substantially zero.

13. The method according to claim 11, wherein said e) allows said d) to continue until the number of samples each of whose classification results do not match decreases to substantially zero.

14. The method according to claim 11, wherein said initial discriminant function and said new discriminant function are each generated by performing feature extraction on an initial parameter set pregenerated for said sample set prepared as said training data, thereby forming a final parameter set, and by performing said discriminant analysis using said final parameter set.

15. An apparatus for generating a two-class classification/prediction model, comprising:

an input device which enters as training data a sample set that contains a plurality of samples belonging to a first class and a plurality of samples belonging to a second class;

a discriminant function generating device which generates, by performing discriminant analysis on said sample set, a first discriminant function having a high classification characteristic for said first class and a second discriminant function having a high classification characteristic for said second class;

a classification result comparing device which classifies said sample set by using said first and second discriminant functions, and isolates any sample whose classification results by said first and second discriminant functions do not match; and

a control device which forms a new sample set by grouping together any sample isolated by said classification result comparing device, and causes said discriminant function generating device and said classification result comparing device to operate repeatedly, and wherein

said control device causes said repeating operation to stop when the number of samples each of whose classification results do not match in said classification result comparing device has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value.

16. The apparatus according to claim 15, wherein said discriminant function generating device generates each of said first and second discriminant functions by performing feature extraction on an initial parameter set pregenerated for said sample set prepared as said training data, thereby forming a final parameter set, and by performing said discriminant analysis using said final parameter set.

17. An apparatus for generating a chemical toxicity prediction model, comprising:

an input device which enters as training data a sample set that contains a plurality of chemicals belonging to a first class and a plurality of chemicals belonging to a second class, wherein said chemicals in said first class have a specific kind of toxicity and said chemicals in said second class do not have said toxicity;

a discriminant function generating device which generates, by performing discriminant analysis on said sample set, a first discriminant function having a high classification characteristic for said first class and a second discriminant function having a high classification characteristic for said second class;

a classification result comparing device which classifies said sample set by using said first and second discriminant functions, and isolates any chemical whose classification results by said first and second discriminant functions do not match; and

a control device which forms a new sample set by grouping together any chemical isolated by said classification result comparing device, and causes said discriminant function generating device and said classification result comparing device to operate repeatedly, and wherein

said control device causes said repeating operation to stop when the number of chemicals each of whose classification results do not match in said classification result comparing device has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value.

18. The apparatus according to claim 17, wherein said control device causes said repeating operation to stop when the number of chemicals each of whose classification results do not match has decreased to substantially zero.

19. The apparatus according to claim 17, wherein said discriminant function generating device generates each of said first and second discriminant functions by performing feature extraction on an initial parameter set pregenerated for said sample set prepared as said training data, thereby forming a final parameter set, and by performing said discriminant analysis using said final parameter set.

20. A classification/prediction method for samples of unknown classes, comprising:

generating a plurality of discriminant function sets each containing a first discriminant function and a second discriminant function, wherein said plurality of discriminant function sets are each generated by carrying out

a) preparing as training data a sample set that contains a plurality of samples belonging to a first class and a plurality of samples belonging to a second class,

b) generating, by performing discriminant analysis on said sample set, said first discriminant function which has a high classification characteristic for said first class and said second discriminant function which has a high classification characteristic for said second class;

c) by classifying said sample set using said first and second discriminant functions, isolating any sample whose classification results by said first and second discriminant functions do not match;

d) repeating said b) and c)by using a new sample set which is formed by grouping together any sample isolated in said c); and

e) causing said d) to stop when the number of samples each of whose classification results do not match in said c) has decreased to or below a predetermined value or when the number of repetitions or processing time for repetitions has reached or exceeded a predetermined value;

obtaining classification results by applying said first and second discriminant functions contained in a first generated one of said plurality of discriminant function sets to said samples of unknown classes; and

sequentially applying said discriminant function sets in order of generation to said samples of unknown classes until said obtained classification results match, wherein

the class indicated by said classification results when said classification results match is predicted to be the class to which a corresponding one of said samples of unknown classes belongs.