System and process for a neural network classification for insurance underwriting suitable for use by an automated system

• US 20040236611A1
• Filed: 04/30/2003
• Published: 11/25/2004
• Est. Priority Date: 04/30/2003
• Status: Abandoned Application

• Current Assignee: Pacific Life Insurance Company (Pacific Mutual Holding Company)
• Original Assignee: GE Financial Assurance Holdings Incorporated (General Electric Company)
• Inventors: Bonissone, Piero Patrone, Yan, Weizhong, Subbu, Rajesh Venkat, Chakraborty, Anindya
• Application Number:
  US10/425,610
• Publication Number:
  US 20040236611A1
• Time in Patent Office: Days
• US Class Current: 705/4
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: G06Q 40/08 Insurance

Abstract

A method and system for designing a neural network classifier and such a neural network for an automated insurance underwriting system and/or its quality assurance is described. While the design method is demonstrated for quality assurance of automated insurance underwriting, it is broadly applicable to diverse decision-making applications in business, commercial, and manufacturing processes. Specifically, multi-class classification problems are solved by decomposing a multi-class classifier into multiple binary-classifiers, which reduces the complexity of the neural network structure, thus reducing the training time and improving the classification performance. Furthermore, the invention also describes a method to incorporate the domain knowledge into the neural network classifier. Both methods work to improve the performance of the classifier.

Claims

1. A system for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the system comprising:

a processing module, where the processing module;

a) processes a plurality of features associated with the insurance application; and

b) outputs a processed result for each of the plurality of features;

a plurality of classifiers, where there is a classifier for each classification and where each of the classifiers;

a) receives the plurality of processed results; and

b) outputs a classification result for the insurance application;

at least one post-processor module, where the at least one post-processing module;

a) processes each of the classification results; and

b) outputs a processed classification result; and

a classification selector for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

2. The system according to claim 1, where the plurality of classifiers are binary classifiers.

3. The system according to claim 2, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

4. The system according to claim 1, where the processing performed by the pre-processing module comprises at least one of:

a) range normalization; and

b) feature extraction.

5. The system according to claim 1, where the processing performed by the pre-processing module comprises generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features.

6. The system according to claim 1, where the processing performed by the post-processing module comprises at least one of:

a) determining cardinality for the plurality of classification results;

b) determining entropy for the plurality of classification results;

c) determining the difference between the highest and second highest values of the classification results; and

d) determining the separation between the rank orders of the highest and second highest values of the classification results.

7. The system according to claim 1, where the selector module is a maximum classification selector which selects the highest of the processed classification results.

14. The process according to claim 1, where processing a plurality of features further comprises at least one of:

a) range normalization; and

b) feature extraction.

8. A system for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the system comprising:

a processing module, where the processing module;

a) processes a plurality of features associated with the insurance application, where the processing performed by the pre-processing module comprises at least one of;

i) range normalization; and

ii) generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features; and

b) outputs a processed result for each of the plurality of features;

a plurality of binary classifiers, where there is a binary classifier for each classification and where each of the binary classifiers;

a) receives the plurality of processed results; and

b) outputs a classification result for the insurance application;

at least one post-processor module, where the at least one post-processing module;

a) processes each of the classification results; and

b) outputs a processed classification result; and

a classification selector for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

9. The system according to claim 8, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

10. The system according to claim 8, where the processing performed by post-processing module comprises at least one of:

a) determining cardinality for the plurality of classification results;

b) determining entropy for the plurality of classification results;

c) determining the difference between the highest and second highest values of the classification results; and

d) determining the separation between the rank orders of the highest and second highest values of the classification results.

11. A process for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the process comprising:

receiving the insurance application, where the insurance application has a plurality of features;

processing a plurality of features associated with the insurance application;

generating a processed result for each of the plurality of features;

generating a classification result for the insurance application based on the processed result;

processing the classification results;

generating a processed classification result; and

selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

12. The process according to claim 11, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification.

13. The process according to claim 12, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

15. The process according to claim 11, where processing the classification results further comprises generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features.

16. The process according to claim 11, where processing the classification results further comprises at least one of:

a) determining cardinality for the plurality of classification results;

b) determining entropy for the plurality of classification results;

c) determining the difference between the highest and second highest values of the classification results; and

d) determining the separation between the rank orders of the highest and second highest values of the classification results.

17. The process according to claim 11, where selecting one of the processed classification results further comprises a maximum classification selector which selects the highest of the processed classification results.

18. A process for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the process comprising:

receiving the insurance application, where the insurance application has a plurality of features;

processing a plurality of features associated with the insurance application, where processing a plurality of features further comprises at least one of;

a) range normalization; and

b) generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features;

generating a processed result for each of the plurality of features;

generating a classification result for the insurance application based on the processed result, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification;

processing the classification results;

generating a processed classification result; and

selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

19. The process according to claim 18, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

20. The process according to claim 18, where processing the classification results further comprises at least one of:

a) determining cardinality for the plurality of classification results;

b) determining entropy for the plurality of classification results;

c) determining the difference between the highest and second highest values of the classification results; and

d) determining the separation between the rank orders of the highest and second highest values of the classification results.

21. A system for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the system comprising:

means for receiving the insurance application, where the insurance application has a plurality of features;

means for processing a plurality of features associated with the insurance application;

means for generating a processed result for each of the plurality of features;

means for generating a classification result for the insurance application based on the processed result;

means for processing the classification results;

means for generating a processed classification result; and

means for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

22. The system according to claim 21, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification.

23. The system according to claim 22, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

24. The system according to claim 21, where the means for processing a plurality of features further comprises at least one of:

a) range normalization; and

b) feature extraction.

25. The system according to claim 21, where the processing by the means for processing the classification results further comprises means for generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features.

26. The system according to claim 21, where the means for processing the classification results further comprises at least one of:

a) means for determining cardinality for the plurality of classification results;

b) means for determining entropy for the plurality of classification results;

c) means for determining the difference between the highest and second highest values of the classification results; and

d) means for determining the separation between the rank orders of the highest and second highest values of the classification results.

27. The system according to claim 21, where the means for selecting is a maximum classification selector which selects the highest of the processed classification results.

28. A system for underwriting an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the system comprising:

means for receiving the insurance application, where the insurance application has a plurality of features;

means for processing a plurality of features associated with the insurance application, where processing a plurality of features further comprises at least one of;

a) range normalization; and

b) generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features;

means for generating a processed result for each of the plurality of features;

means for generating a classification result for the insurance application based on the processed result, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification;

means for processing the classification results;

means for generating a processed classification result; and

means for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

29. The system according to claim 28, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

30. The system according to claim 28, where the means for processing the classification results further comprises at least one of:

a) means for determining cardinality for the plurality of classification results;

b) means for determining entropy for the plurality of classification results;

c) means for determining the difference between the highest and second highest values of the classification results; and

d) means for determining the separation between the rank orders of the highest and second highest values of the classification results.

31. A computer readable medium having code for causing a processor to underwrite an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the computer medium comprising:

code for receiving the insurance application, where the insurance application has a plurality of features;

code for processing a plurality of features associated with the insurance application;

code for generating a processed result for each of the plurality of features;

code for generating a classification result for the insurance application based on the processed result;

code for processing the classification results;

code for generating a processed classification result; and

code for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

32. The medium according to claim 31, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification.

33. The medium according to claim 32, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

34. The medium according to claim 31, where the code for processing a plurality of features further comprises at least one of:

a) range normalization; and

b) feature extraction.

35. The medium according to claim 31, where the code for processing the classification results further comprises code for generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features.

36. The medium according to claim 31, where the code for processing the classification results further comprises at least one of:

a) code for determining cardinality for the plurality of classification results;

b) code for determining entropy for the plurality of classification results;

c) code for determining the difference between the highest and second highest values of the classification results; and

d) code for determining the separation between the rank orders of the highest and second highest values of the classification results.

37. The medium according to claim 31, where the code for selecting is a maximum classification selector which selects the highest of the processed classification results.

38. A computer medium having code for causing a processor to underwrite an insurance application based on a plurality of previous insurance application underwriting decisions, where underwriting the insurance application includes assigning a classification to the insurance application, the medium comprising:

code for receiving the insurance application, where the insurance application has a plurality of features;

code for processing a plurality of features associated with the insurance application, where processing a plurality of features further comprises at least one of;

a) range normalization; and

b) generating a tag, wherein the tag is generated based on one or more rules applied to at least one of the plurality of features;

code for generating a processed result for each of the plurality of features;

code for generating a classification result for the insurance application based on the processed result, where the classification results are generated by a plurality of binary classifiers, and where there is a classifier for each classification;

code for processing the classification results;

code for generating a processed classification result; and

code for selecting one of the processed classification results, where the selection indicates a classification assignment for the insurance application and selects the highest of the processed classification results.

39. The medium according to claim 38, where the classification results from each of the plurality of binary classifiers comprise one of:

a) the selection of the classification associated with the binary classifier; and

b) a selection of the rest of the classifications associated with the other binary classifiers.

40. The medium according to claim 38, where the code for processing the classification results further comprises at least one of:

a) means for determining cardinality for the plurality of classification results;

b) means for determining entropy for the plurality of classification results;

c) means for determining the difference between the highest and second highest values of the classification results; and

d) means for determining the separation between the rank orders of the highest and second highest values of the classification results.