Method of operating artificial intelligence machines to improve predictive model training and performance
View Patent ↗A method of improving the training and performance of predictive models. A first method of operating an artificial intelligence machine produces predictive model language documents describing improved predictive models that generate better business decisions from raw data record inputs. A second method of operating an artificial intelligence machine including processors for predictive model algorithms produces and outputs better business decisions from raw data record inputs. Both methods enrich the raw data records their processors are fed by deleting data fields with data values that have little benefit in decision making, and that derive and add new data fields from information sources then available that do benefit in the decision making of the artificial intelligence machine through improved accuracies of prediction.
1. A computer-implemented method for generating production environment predictive model language documents describing a predictive model and a complementary data-enrichment descriptor, comprising, via one or more processors:
receiving a training data set comprising a first plurality of records reflecting behavior of a plurality of entities, each of the first plurality of records including a first set of data fields containing data values;
analysing the data values of the first plurality of data fields of the training data set against a criterion;
deleting a subset of the first plurality of data fields containing data values that meet the criterion to create a second plurality of records comprising a second set of data fields;
adding a new derivative data field to the second set of data fields to create a third plurality of records comprising a third set of data fields;
training a predictive model on the third plurality of records to create a first trained model, the predictive model comprising a plurality of smart agent profiles and at least one artificial intelligence classifier, each of the plurality of trained smart agent profiles of the first trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the first trained model being trained on records of the third plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
training the predictive model on the first plurality of records to create a second trained model, each of the plurality of trained smart agent profiles of the second trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the second trained model being trained on records of the first plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
inputting a first testing data set comprising a first plurality of testing records having data fields corresponding to the first set of data fields to the second trained model to generate output comprising a first plurality of corresponding predictions;
inputting a second testing data set comprising a second plurality of testing records having data fields corresponding to the third set of data fields to the first trained model to generate output comprising a second plurality of corresponding predictions;
determining that the second plurality of predictions is more accurate than the first plurality of predictions;
generating at least one predictive model language document embodying a data-enrichment descriptor identifying the deleted subset of the first plurality of data fields and the new derivative data field;
generating at least one predictive model language document embodying the first trained model; and
outputting the predictive model language documents embodying the data-enrichment descriptor and the first trained model for use in making predictions in a real-time production environment.
2. The computer-implemented method of claim 1 , wherein the criterion is one of: (a) a threshold number of random data values; (b) a threshold number of repeating data values; and (c) a threshold for Shannon entropy exhibited by data values.
3. The computer-implemented method of claim 1 , wherein the adding of the new derivative data field is contingent on a determination that the second set of data fields is insufficient to support accurate prediction.
4. The computer-implemented method of claim 1 , further comprising, via the one or more processors, initializing the new derivative data field for each of the third plurality of records by performing one of the following: (a) converting a real scaler numeric data value into a fuzzy value; or (b) replacing a symbolic value with a behaviour group data value.
5. The computer-implemented method of claim 1 , wherein the new derivative data field embodies a combination of two or more of the second set of data fields.
6. The computer-implemented method of claim 1 , further comprising, via the one or more processors—
inputting the first testing data set to the first trained model to generate output comprising a third plurality of corresponding predictions;
determining that the second plurality of predictions is more accurate than the third plurality of predictions.
7. The computer-implemented method of claim 1 , further comprising, via the one or more processors—
implementing the predictive model language documents embodying the data-enrichment descriptor and the first trained model in a real-time production environment at least in part by—
receiving a production set of a records, each of the production set of records comprising a plurality of production data fields,
modifying the plurality of production data fields according to the at least one predictive model language document embodying the data-enrichment descriptor to create a modified production set of records,
instantiating the first trained model according to the at least one predictive model language document embodying the first trained model,
inputting the modified production set of records to the first trained model to generate a third plurality of corresponding predictions.
8. At least one server for generating production environment predictive model language documents describing a predictive model and a complementary data-enrichment descriptor, the at least one server comprising:
one or more processors;
non-transitory computer-readable storage media having computer-executable instructions stored thereon, wherein when executed by the one or more processors the computer-readable instructions cause the one or more processors to—
receive a training data set comprising a first plurality of records reflecting behavior of a plurality of entities, each of the first plurality of records including a first set of data fields containing data values;
analyse the data values of the first plurality of data fields of the training data set against a criterion;
delete a subset of the first plurality of data fields containing data values that meet the criterion to create a second plurality of records comprising a second set of data fields;
add a new derivative data field to the second set of data fields to create a third plurality of records comprising a third set of data fields;
train a predictive model on the third plurality of records to create a first trained model, the predictive model comprising a plurality of smart agent profiles and at least one artificial intelligence classifier, each of the plurality of trained smart agent profiles of the first trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the first trained model being trained on records of the third plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
train the predictive model on the first plurality of records to create a second trained model, each of the plurality of trained smart agent profiles of the second trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the second trained model being trained on records of the first plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
input a first testing data set comprising a first plurality of testing records having data fields corresponding to the first set of data fields to the second trained model to generate output comprising a first plurality of corresponding predictions;
input a second testing data set comprising a second plurality of testing records having data fields corresponding to the third set of data fields to the first trained model to generate output comprising a second plurality of corresponding predictions;
determine that the second plurality of predictions is more accurate than the first plurality of predictions;
generate at least one predictive model language document embodying a data-enrichment descriptor identifying the deleted subset of the first plurality of data fields and the new derivative data field;
generate at least one predictive model language document embodying the first trained model; and
output the predictive model language documents embodying the data-enrichment descriptor and the first trained model for use in making predictions in a real-time production environment.
9. The at least one server of claim 8 , wherein the criterion is one of: (a) a threshold number of random data values; (b) a threshold number of repeating data values; and (c) a threshold for Shannon entropy exhibited by data values.
10. The at least one server of claim 8 , wherein the adding of the new derivative data field is contingent on a determination that the second set of data fields is insufficient to support accurate prediction.
11. The at least one server of claim 8 , wherein when executed by the one or more processors the computer-readable instructions further cause the one or more processors to, initialize the new derivative data field for each of the third plurality of records by performing one of the following: (a) converting a real scaler numeric data value into a fuzzy value; or (b) replacing a symbolic value with a behaviour group data value.
12. The at least one server of claim 8 , wherein the new derivative data field embodies a combination of two or more of the second set of data fields.
13. The at least one server of claim 8 , wherein when executed by the one or more processors the computer-readable instructions further cause the one or more processors to—
input the first testing data set to the first trained model to generate output comprising a third plurality of corresponding predictions;
determine that the second plurality of predictions is more accurate than the third plurality of predictions.
14. The at least one server of claim 8 , wherein when executed by the one or more processors the computer-readable instructions further cause the one or more processors to—
implement the predictive model language documents embodying the data-enrichment descriptor and the first trained model in a real-time production environment at least in part by—
receiving a production set of a records, each of the production set of records comprising a plurality of production data fields,
modifying the plurality of production data fields according to the at least one predictive model language document embodying the data-enrichment descriptor to create a modified production set of records,
instantiating the first trained model according to the at least one predictive model language document embodying the first trained model,
inputting the modified production set of records to the first trained model to generate a third plurality of corresponding predictions.
15. Non-transitory computer-readable storage media having computer-executable instructions for generating production environment predictive model language documents describing a predictive model and a complementary data-enrichment descriptor, wherein when executed by at least one processor the computer-readable instructions cause the at least one processor to:
receive a training data set comprising a first plurality of records reflecting behavior of a plurality of entities, each of the first plurality of records including a first set of data fields containing data values;
analyse the data values of the first plurality of data fields of the training data set against a criterion;
delete a subset of the first plurality of data fields containing data values that meet the criterion to create a second plurality of records comprising a second set of data fields;
add a new derivative data field to the second set of data fields to create a third plurality of records comprising a third set of data fields;
train a predictive model on the third plurality of records to create a first trained model, the predictive model comprising a plurality of smart agent profiles and at least one artificial intelligence classifier, each of the plurality of trained smart agent profiles of the first trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the first trained model being trained on records of the third plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
train the predictive model on the first plurality of records to create a second trained model, each of the plurality of trained smart agent profiles of the second trained model comprising a representation of historical data reflecting behavior of one of the plurality of entities, the at least one classifier of the second trained model being trained on records of the first plurality of records reflecting behavior of at least two of the plurality of entities and being constructed according to one of: a neural network, case based reasoning, a decision tree, a genetic algorithm, fuzzy logic, and rules and constraints;
input a first testing data set comprising a first plurality of testing records having data fields corresponding to the first set of data fields to the second trained model to generate output comprising a first plurality of corresponding predictions;
input a second testing data set comprising a second plurality of testing records having data fields corresponding to the third set of data fields to the first trained model to generate output comprising a second plurality of corresponding predictions;
determine that the second plurality of predictions is more accurate than the first plurality of predictions;
generate at least one predictive model language document embodying a data-enrichment descriptor identifying the deleted subset of the first plurality of data fields and the new derivative data field;
generate at least one predictive model language document embodying the first trained model; and
output the predictive model language documents embodying the data-enrichment descriptor and the first trained model for use in making predictions in a real-time production environment.
16. The computer-readable storage media of claim 15 , wherein the criterion is one of: (a) a threshold number of random data values; (b) a threshold number of repeating data values; and (c) a threshold for Shannon entropy exhibited by data values.
17. The computer-readable storage media of claim 15 , wherein when executed by the at least one processor the computer-readable instructions further cause the at least one processor to, initialize the new derivative data field for each of the third plurality of records by performing one of the following: (a) converting a real scaler numeric data value into a fuzzy value; or (b) replacing a symbolic value with a behaviour group data value.
18. The computer-readable storage media of claim 15 , wherein the new derivative data field embodies a combination of two or more of the second set of data fields.
19. The computer-readable storage media of claim 15 , wherein when executed by the at least one processor the computer-readable instructions further cause the at least one processor to—
input the first testing data set to the first trained model to generate output comprising a third plurality of corresponding predictions;
determine that the second plurality of predictions is more accurate than the third plurality of predictions.
20. The computer-readable storage media of claim 15 , wherein when executed by the at least one processor the computer-readable instructions further cause the at least one processor to—
implement the predictive model language documents embodying the data-enrichment descriptor and the first trained model in a real-time production environment at least in part by—
receiving a production set of a records, each of the production set of records comprising a plurality of production data fields,
modifying the plurality of production data fields according to the at least one predictive model language document embodying the data-enrichment descriptor to create a modified production set of records,
instantiating the first trained model according to the at least one predictive model language document embodying the first trained model,
inputting the modified production set of records to the first trained model to generate a third plurality of corresponding predictions.