Method and apparatus for inducing rules from data classifiers

• US 6,336,109 B2
• Filed: 04/15/1997
• Issued: 01/01/2002
• Est. Priority Date: 04/15/1997
• Status: Expired due to Fees

• Current Assignee: Bell Northern Research, LLC (Hilco Inc. (d/b/a Hilco Global)), Cerebrus Solutions Ltd (Neural Technologies Limited)
• Original Assignee: Cerebrus Solutions Ltd (Neural Technologies Limited)
• Inventors: Howard, Gary
• Primary Examiner(s): Powell, Mark R.
• Assistant Examiner(s): Starks, Wilbert L.
• Application Number:
  US08/840,115
• Publication Number:
  US 20010011259A1
• Time in Patent Office: 1,722 Days
• Field of Search: 706/2, 706/12, 706/13, 706/25, 706/84, 706/41, 370/330, 382/229
• US Class Current: 706/25
• CPC Class Codes: G06F 18/24765 Rule-based classification
• CPC Class Codes: G06N 3/042 Knowledge-based neural networks; Logical representations of neural networks
• CPC Class Codes: G06N 5/025 Extracting rules from data
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A method of processing data relating to a plurality of examples using a data classifier arranged to classify input data into one of a number of classes, and a rule inducer, comprising the steps of:

(i) inputting a series of inputs to the data classifier so as to obtain a series of corresponding outputs;

(ii) inputting said series of outputs and at least some of said series of inputs to the rule inducer so as to obtain a series of rules which describe relationships between the series of inputs to the data classifier and the series of corresponding outputs from the data classifier.

Claims

1. A method of using a computer to analyze data representative of subscriber account usage comprising steps of:

(i) providing a data classifier arranged to receive a series of inputs comprising the data representative of subscriber account usage, and to produce responsive thereto classification outputs indicative of which one of a number of classes each of said series of inputs is a member of;

(ii) providing a rule inducer arranged to receive at least some of the series of inputs and to receive from the data classifier only the classification outputs;

(iii) inputting the series of inputs to the data classifier so as to obtain a series of corresponding outputs each comprising an indication of which one of the number of classes the input is a member of;

(vi) operating the rule inducer to produce, responsive to the at least some of the series of inputs and the classification outputs, a series of rules descriptive of relationships between the series of inputs and the classification outputs.

2. A method as claimed in claim 1 wherein the data classifier is unsupervised.

3. A method as claimed in claim 1 wherein the data classifier comprises a neural network.

4. A method as claimed in claim 1 wherein the data classifier comprises a Bayesian classifier.

5. A method as claimed in claim 1 wherein said step (ii) comprises combining each input with its corresponding output so as to create a second series of inputs;

and inputting the second series of inputs to the rule inducer.

6. A method as claimed in claim 1 which further comprises the step of transforming the series of rules into a format such that the formatted rules can be used as a data classifier.

7. A method as claimed in claim 1 which further comprises the step of:

incorporating the rules in to a case-based reasoning system such that in use the case-based reasoning system can be used as a data classifier.

9. A method as claimed in claim 1 wherein the data classified is arranged to classify input data into one of a number of classes according to whether an input is indicative of a fraud or not.

10. A method as claimed in claim 9 wherein said fraud relates to fraudulent use of a telecommunications system.

16. Apparatus according to claim 5 wherein said data classifier comprises a neural net.

17. A method according to claim 1 in which the subscriber account is a telecommunication subscriber account.

18. A method according to claim 1 in which the subscriber account is a credit card subscriber account.

8. A computer system for analyzing data representative of subscriber account usage comprising:

a data classifier arranged to receive a series of inputs comprising the data representative of subscriber account usage, and to produce responsive thereto classification outputs each comprising an indication of which one of a number of classes each of said inputs is a member of;

a rule inducer arranged to receive at least some of said series of inputs to said data classifier and to receive from said data classifier only said series of outputs, and to produce, responsive to said at least some of said series of inputs and said series of outputs, a series of rules descriptive of relationships between said series of inputs and said classification outputs.

11. A method of processing data representative of subscriber account usage relating to a plurality of examples using a data classifier arranged to classify input data into one of a number of classes, and a rule inducer, comprising the steps of:

(i) inputting a series of inputs to the data classifier so as to obtain a series of corresponding outputs each comprising an indication of which of the number of classes the input is a member of;

(ii) inputting said series of outputs and at least some of said series of inputs to the rule inducer so as to obtain a series of rules which describe relationships between the series of inputs to the data classifier and the series of corresponding outputs from the data classifier;

wherein said series of rules is obtained independent of information internal to the data classifier.

12. A method of processing data representative of subscriber account usage relating to a plurality of examples using a data classifier arranged to classify input data into one of a number of classes, and a rule inducer, comprising the steps of:

(i) inputting a series of inputs to the data classifier so as to obtain a series of corresponding outputs each comprising an indication of which of the number of classes the input is a member of;

(ii) inputting said series of outputs and at least some of said series of inputs to the rule inducer so as to obtain a series of rules which describe relationships between the series of inputs to the data classifier and the series of corresponding outputs from the data classifier;

wherein said series of rules is obtained independent of internal structure of said data classifier.

13. A method of identifying characteristics of a class of fraudulent use of a telecommunications network customer account, the method comprising;

using a data classifier arranged to classify input customer account records representative of subscriber account usage into one of a number of classes indicative of fraud and non-fraud candidate data;

inducing rules from customer account records input to the data classifier and from the classification information output from the data classifier, said output information comprising a set of attributes for each customer account and a class assignment for that customer account; and

generating rules from the rule inducer so as to obtain a series of rules which describe relationships between the customer account records input to the data classifier and the corresponding outputs from the data classifier and wherein the rules characterise those classes to which candidate fraud data has been assigned.

14. A method according to claim 13 wherein said data classifier comprises a neural net.

15. Apparatus for identifying and characterising anomalous behaviour relating to potential fraudulent use of a network, the apparatus comprising;

a data classifier arranged to classify input call detail records representative of subscriber account usage into one of a number of classes indicative of fraud and non-fraud candidate data;

and a rule inducer for inducing rules from call detail record data input to the data classifier and from information output from the data classifier, said output information comprising a set of attributes for each customer account and a class assignment for that customer account;

said rule inducer being arranged to generate rules induced therein so as to obtain a series of rules which characterise relationships between the call detail record data input to the data classifier and the series of corresponding outputs from the data classifier whereby to provide a characterisation of those classes to which the candidate fraud data has been assigned.