Predictive outputs in the presence of entity-dependent classes

1. A method of improving quality of predictive model outputs, the method comprising:

receiving information describing a class association function that generates input data from source data, wherein the source data comprises a plurality of entities of one or more entity types, wherein the input data comprises a combination of the source data with class information such that entities in the input data are associated with one or more classes, and wherein the input data is configured to be input to a predictive model as data on which prediction is performed;

deriving, from the information describing the class association function, class association function characteristic data comprising inferences relating to operation of the class association function that are not derivable solely from the input data;

constructing a class-specific transformation function that transforms the input data into improved input data; and

transforming, using a computer processor, the input data into the improved input data using the class-specific transformation function and the class association function characteristic data, wherein the improved input data is configured to be input into the same predictive model as data on which prediction is performed, and wherein the improved input data is not based on output of the predictive model generated using the input data as input to the predictive model.

2. The method of claim 1 , further comprising generating, using the computer processor, the input data by providing the source data as input to the class association function and receiving the input data as output from the class association function.

3. The method of claim 1 , wherein the source data comprises data stored in computer memory, a database, data warehouse, data lake, data lakehouse, file system or object store.

4. The method of claim 1 , wherein the information describing the class association function comprises a class association function that is directly introspectable.

5. The method of claim 1 , wherein the information describing the class association function comprises a class association function that is not directly introspectable.

6. The method of claim 1 , wherein the information describing the class association function is received from an external source.

7. The method of claim 1 , wherein receiving the information describing the class association function comprises generating the information describing the class association function by approximating at least a portion of the class association function using information inferred from the source data and the class information.

8. The method of claim 1 , wherein the deriving comprises one or more of:

(i) extracting entity-related information from the class association function;

(ii) applying the class association to one or more subsets of the input data to identify entity-level attributes; and

(iii) analyzing the information describing the class association function to determine an approximate implementation of the class association function.

9. The method of claim 1 , wherein the information describing the class association function is incorporated into the class-specific transformation function, such that the class-specific transformation function generates the input data from the source data.

10. The method of claim 1 , wherein the information describing the class association function information is either incorporated into the class-specific transformation function or used as in input to the class-specific transformation function.

11. The method of claim 1 , further comprising providing the improved input data as input to the predictive model and predicting, using the predictive model, whether a particular entity will be assigned to a particular class.

12. The method of claim 1 , wherein the improved input data, when used as input to the predictive model, results in the predictive model providing output of greater predictive accuracy compared to use of the input data as input to the predictive model.

13. A method of improving quality of predictive model outputs, the method comprising:

receiving class association function characteristic data derived from information describing a class association function that generates input data from source data, wherein the source data comprises a plurality of entities of one or more entity types, wherein the input data comprises a combination of the source data with class information such that entities in the input data are associated with one or more classes, wherein the input data is configured to be input to a predictive model as data on which prediction is performed, and wherein the class association function characteristic data comprises inferences relating to operation of the class association function that are not derivable solely from the input data;

constructing a class-specific transformation function that transforms the input data into improved input data; and

transforming, using a computer processor, the input data into the improved input data using the class-specific transformation function and the class association function characteristic data, wherein the improved input data is configured to be input into the same predictive model as data on which prediction is performed, and wherein the improved input data is not based on output of the predictive model generated using the input data as input to the predictive model.

14. A method of improving quality of predictive model outputs for a healthcare system, the method comprising:

receiving information describing a class association function that generates input data from source data, wherein the source data comprises healthcare information associated with a plurality of participants associated with a healthcare system, wherein the input data comprises a combination of the source data with class information such that participants associated with the healthcare system in the input data are associated with one or more classes, and wherein the input data is configured to be input to a predictive model as data on which prediction is performed;

deriving, from the information describing the class association function, class association function characteristic data comprising inferences relating to operation of the class association function that are not derivable solely from the input data;

constructing a class-specific transformation function that transforms the input data into improved input data;

transforming, using a computer processor, the input data into the improved input data using the class-specific transformation function and the class association function characteristic data, wherein the improved input data is configured to be input into the same predictive model as data on which prediction is performed, and wherein the improved input data is not based on output of the predictive model generated using the input data as input to the predictive model; and

providing the improved input data as input to the predictive model and predicting, using the predictive model and the improved input data, whether a particular participant associated with the healthcare system will satisfy a healthcare-related condition.

15. The method of claim 14 , wherein the predicting comprises predicting, using the predictive model, whether the particular participant associated with the healthcare system will satisfy a predicate over data associated with the particular participant.

16. The method of claim 15 , wherein the predicate comprises a patient receiving a diagnosis, a patient receiving a procedure, a patient receiving a treatment, a healthcare professional making a diagnosis, a healthcare professional performing a procedure, a healthcare professional prescribing a treatment, a payer making a payment, a payer rejecting a claim, or any combination of the foregoing.

17. The method of claim 14 , wherein deriving the class association function characteristic data comprises determining that entities are assigned to the one or more classes based on a first date associated with a first predicate over data associated with the entities.

18. The method of claim 17 , wherein deriving the class association function characteristic data comprises determining that entities are assigned to the one or more classes based on a relation of (i) the first date associated with the first predicate to (ii) a second date associated with a second predicate over the data associated with the entities.

19. The method of claim 18 , wherein transforming the input data into the improved input data comprises filtering the input data to remove information associated with at least one of (i) dates temporally following an earliest date of the second predicate and (ii) dates temporally preceding an earliest date of the second predicate.

20. The method of claim 14 , wherein deriving the class association function characteristic data comprises determining that entities are assigned to the one or more classes based on an occurrence of a healthcare-related event comprising starting a therapy, receiving a treatment, receiving a procedure, or receiving a diagnosis.

21. The method of claim 20 , wherein transforming the input data into the improved input data comprises filtering the input data to remove information associated with dates having a particular temporal relationship to a time or date associated with the healthcare-related event.

22. The method of claim 21 , wherein the particular temporal relationship comprises an earlier time or data, a later time or date, or a same time or date.

23. The method of claim 14 , wherein the participants associated with the healthcare system comprise at least one of patients, healthcare providers, healthcare-related organizations, and payers.

24. A system for improving quality of predictive model outputs, the system comprising:

a processor and a memory storing computer-executable instructions that, when executed by the processor, program the processor to perform the operation of transforming input data into improved input data using a class-specific transformation function and class association function characteristic data, wherein the improved input data is configured to be input into a predictive model as data on which prediction is performed, and wherein the improved input data is not based on output of the predictive model generated using the input data as input to the predictive model;

wherein the input data comprises a combination of source data with class information such that entities in the input data are associated with one or more classes, wherein the input data is configured to be input to the same predictive model as data on which prediction is performed, wherein the source data comprises a plurality of entities of one or more entity types;

wherein the class-specific transformation function is constructed to transform the input data into the improved input data; and

wherein the class association function characteristic data is derived from information describing a class association function, wherein the class association function generates the input data from the source data, and wherein the class association function characteristic data comprises inferences relating to operation of the class association function that are not derivable solely from the input data.

25. The system of claim 24 , wherein the processor further performs the operation of generating the input data by providing the source data as input to the class association function and receiving the input data as output from the class association function.

26. The system of claim 24 , wherein the information describing the class association function comprises a class association function that is directly introspectable.

27. The system of claim 24 , wherein the information describing the class association function comprises a class association function that is not directly introspectable.

28. The system of claim 24 , wherein the information describing the class association function is generated by approximating at least a portion of the class association function using information inferred from the source data and the class information.

29. The system of claim 24 , wherein the class association function characteristic data is derived by one or more of:

(i) extracting entity-related information from the class association function;

(ii) applying the class association to one or more subsets of the input data to identify entity-level attributes; and

(iii) analyzing the information describing the class association function to determine an approximate implementation of the class association function.

30. The system of claim 24 , wherein the processor further performs the operation of providing the improved input data as input to the predictive model and predicting, using the predictive model, whether a particular entity will be assigned to a particular class.