Anomaly analysis using maximum likelihood continuity mapping

• US 6,038,388 A
• Filed: 01/29/1998
• Issued: 03/14/2000
• Est. Priority Date: 01/30/1997
• Status: Expired due to Fees

• Current Assignee: Los Alamos National Security LLC (Government of the United States of America)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: White, James M., Scovel, James Clinton, Hogden, John E.
• Primary Examiner(s): Teska, Kevin J.
• Assistant Examiner(s): JONES, HUGH M
• Application Number:
  US09/015,093
• Time in Patent Office: 775 Days
• Field of Search: 395/500.23, 395/500.27, 705/2, 705/3, 705/11
• US Class Current: 703/6
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04

Abstract

A method of modeling symbol sequence generation is obtained which allows the probability of symbol sequences to be estimated. The model involves treating symbols as if they are emitted as a point travels through an abstract space called a continuity map (CM), in which each position in the space has associated probabilities of emitting each of the possible symbols. This method for modeling symbol generation, combined with methods for estimating the probability of symbol sequences given smooth paths through the CM, can be applied to such problems as language modeling and anomaly/fraud detection. A fraud detection study is described that demonstrates that the invention can be used for detecting medical fraud.

Claims

1. A computer-implemented method for detecting anomalous medical procedures comprising the steps of:

finding a mapping between model medical procedures forming model medical treatment sequences and probability density functions (PDFs) over continuity map (CM) positions in a health space, wherein each PDF quantifies the probability of positions in the CM given the associated model medical procedure;

inputting a new sequence of medical procedures to be reviewed;

finding a path through said continuity map that maximizes the conditional probability of said new sequence of medical procedures; and

identifying anomalous medical procedures as any sequence of medical procedures with a conditional probability of the sequence of medical procedures given the path below a threshold value.

2. A method according to claim 1, wherein said step of finding said mapping further includes the steps of:

(a) forming a first estimate of said mapping between said model medical procedures and locations and shapes of said PDFs on said CM;

(b) determining for each sequence of said model medical procedures a smooth path through said CM that maximizes the probability of said each model medical treatment sequence;

(c) adjusting said locations and said shapes of said PDFs to increase the probability of said each model medical treatment sequence to form a new estimate of said mapping; and

(d) repeating steps (b)-(c) until the difference in said locations and said shapes of said PDFs between successive ones of said mapping is below a selected threshold.

3. A method according to claim 2, wherein the step of determining each said model path on said CM further includes the steps of:

inputting a sequence of model medical procedures forming the model medical treatment sequence;

choosing a trial path on said CM;

adjusting the trial path to meet smoothness constraints on the model medical treatment sequence to get a constrained path;

adjusting said constrained path to maximize the probability of obtaining said model medical treatment sequence.

4. A computer-implemented method for identifying anomalous transactions comprising the steps of:

finding a mapping between model events forming model transactions and probability density functions (PDFs) over continuity map (CM) positions, wherein each PDF quantifies the probability of positions on the CM given the associated model event;

inputting a new event sequence forming a new transaction to be reviewed;

finding a path through said continuity map that maximizes the conditional probability of said new event sequence; and

identifying anomalous transactions as any sequence of events with a conditional probability of the sequence of events below a threshold.

5. A method according to claim 4, wherein said step of finding said mapping further includes the steps of:

(a) forming a first estimate of said mapping between said transaction model events and locations and shapes of said PDFs on said CM;

(b) determining for each sequence of said model events a path through said CM that maximizes the probability of said each transaction;

(c) adjusting said locations and said shapes of said PDFs to increase the probability of said each transaction to form a new estimate of said mapping; and

(d) repeating steps (b)-(c) until the difference in said locations and said shapes of said PDFs between successive ones of said mapping is below a selected threshold.

6. A method according to claim 5, wherein the step of determining each said model path on said CM further includes the steps of:

inputting a model event sequence;

choosing a trial path on said CM;

adjusting the trial path to meet selected smoothness constraints on the transaction to get a constrained path; and

adjusting said constrained path to maximize the probability of obtaining said transaction.