System and method for predicting rare events

• US 8,244,656 B2
• Filed: 09/25/2008
• Issued: 08/14/2012
• Est. Priority Date: 09/25/2008
• Status: Expired due to Fees

• Current Assignee: Air Products and Chemicals Incorporated (Linde Plc)
• Original Assignee: Air Products and Chemicals Incorporated (Linde Plc)
• Inventors: Petit, Cecilia Anna Paulette
• Primary Examiner(s): Vincent, David
• Application Number:
  US12/284,976
• Publication Number:
  US 20100076917A1
• Time in Patent Office: 1,419 Days
• Field of Search: 706/12, 706/45
• US Class Current: 706/45
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y02A 90/10 Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

Systems and methods are provided for predicting rare events, such as hospitalization events. Predictor sequences may be generated by example systems and methods. Further, the fitness of those sequences may be measured. Sequences may be refined and/or combined with other sequences to produce better sequences. Related sequences may have their relationship identified and associated with the respective sequences. Finally, the sequences may be used to create a predictive model designed to determine, based on a sequence of events related to a person, if a hospitalization event is likely to occur in a given timeframe. Other models may be constructed and used to predict other rare events, based on related event sequences.

Claims

1. A computer-implemented method for predicting rare events, comprising:

postulating, by a computer processor, at least one predictor sequence, wherein each predictor sequence includes a plurality of variables, the postulating including;

generating a random value for each variable, wherein the value is within a range associated with the variable;

adding the variable to a predictor sequence; and

repeating the generating and repeating the adding, until one of the following conditions is met;

the plurality of variables represents a specified length of time and

the plurality of variables represents a specified number of events;

for each predictor sequence, determining, by the processor, if the predictor sequence has an adequate fitness, wherein fitness is measured by determining a correlation between the predictor sequence and a rare event to be predicted;

if the predictor sequence does not have an adequate fitness, refining the predictor sequence until the predictor sequence has adequate fitness; and

predicting the occurrence of the rare event based at least in part on a model, wherein the model is constructed from the at least one predictor sequence having adequate fitness.

2. The method of claim 1, wherein the rare event is a medical event.

3. The method of claim 1, wherein the rare event is a hospitalization event.

4. The method of claim 1, wherein the rare event includes at least one of the group:

a machine, failure of equipment, and a plant failure.

5. The method of claim 1, wherein refining includes discarding the current predictor sequence and randomly generating values for a new predictor sequence.

6. The method of claim 1, wherein refining includes randomly adjusting at least one of a variable value and an order of the variables of the predictor sequence.

7. The method of claim 1, wherein refining includes breeding the predictor sequence with another predictor sequence, by randomly combining elements of the predictor sequence with elements of the another predictor sequence to produce a resulting sequence having the same size as the predictor sequence and the another predictor sequence.

8. The method of claim 1, wherein the specified length or specified number of events corresponding to the one of the conditions is a random number generated within a range of possible numbers.

9. The method of claim 8, wherein the range of possible numbers is received from user input.

10. The method of claim 1, further comprising:

storing each predictor sequence in a predictor sequence library.

11. The method of claim 1, further comprising:

postulating at least two predictor sequences;

determining if two or more predictor sequences of the at least two predictor sequences are related, based on a comparison of fitness measurements of the at least two predictor sequences; and

if related, associating, with each related predictor sequence, a rule describing the relation.

12. The method of claim 11, wherein two or more predictor sequences are related when the two or more predictor sequences have similar fitness measurements.

13. The method of claim 11, further comprising:

determining whether two or more predictor sequences, which have identical predictor sequence variable values, except that one variable in one of the predictor sequences differs from the other predictor sequence, are related.

14. The method of claim 11, further comprising:

determining whether two or more predictor sequences, which have an identical composition of variables, but being arranged in a different order for each predictor sequence, are related.

15. The method of claim 11, further comprising:

determining whether two or more predictor sequences, which are identical except for the insertion or deletion of one or more variables, predict similar, but not identical results.

20. The method of claim 1, wherein the correlation is measured as one of a percentage of correctly predicted outcomes, a number of false negatives, and a number of false positives.

21. The method of claim 11, further comprising:

identifying a variable contained in at least one, but not all predictor sequences in a set of two or more related predictor sequences; and

designating the identified variable as a wild card variable.

22. The method of claim 15, further comprising:

associating a penalty with the one or more variables that are inserted or deleted, relative to one of the two or more predictor sequences for which the one or more variables were not inserted or deleted.

16. A system for predicting rare events, comprising:

a module configured to postulate at least one predictor sequence, wherein each predictor sequence includes a plurality of variables, the postulating including;

generating a random value for each variable, wherein the value is within a range associated with the variable;

adding the variable to a predictor sequence; and

repeating the generating and repeating the adding, until one of the following conditions is met;

the plurality of variables represents a specified length of time and

the plurality of variables represents a specified number of events;

the module configured to determine, for each predictor sequence of the at least one predictor sequence, if the predictor sequence has an adequate fitness, wherein fitness is measured by determining a correlation between the predictor sequence and a rare event to be predicted;

the module configured to refine the predictor sequence, if the predictor sequence does not have an adequate fitness, until the predictor sequence has adequate fitness; and

the module configured to predict the occurrence of the rare event based at least in part on a model, wherein the model is constructed from the at least one predictor sequence having adequate fitness.

17. The system of claim 16, wherein the model is constructed by combining multiple predictor sequences using Boolean operators, thereby expressing a probability of an event as a Boolean expression that includes the combined predictor sequences, linked by the Boolean operators.

18. The system of claim 16, wherein the model is constructed by any one of the group including:

building at least one decision tree having at least one predictor sequence as a node, and running at least one Monte Carlo method.

19. A computer-readable storage medium encoded with instructions configured to be executed by a processor, the instructions which, when executed by the processor, cause the performance of the following method:

postulating at least one predictor sequence, wherein each predictor sequence includes a plurality of variables, the postulating including;

generating a random value for each variable, wherein the value is within a range associated with the variable;

adding the variable to a predictor sequence; and

repeating the generating and repeating the adding, until one of the following conditions is met;

the plurality of variables represents a specified length of time and

the plurality of variables represents a specified number of events;

for each predictor sequence, determining if the predictor sequence has an adequate fitness, wherein fitness is measured by determining a correlation between the predictor sequence and a rare event to be predicted;

if the predictor sequence does not have an adequate fitness, refining the predictor sequence until the predictor sequence has adequate fitness; and

predicting the occurrence of the rare event based at least in part on a model, wherein the model is constructed from the at least one predictor sequences having adequate fitness.