Acceptance status classification of product-related data structures using models with multiple training periods
Techniques are provided for acceptance status classification of product-related data structures using models with multiple training periods. One method comprises obtaining data for a given product-related data structure; evaluating first features related to the given product-related data structure using the obtained data; applying the first features related to the given product-related data structure to one or more models trained using multiple different training periods to obtain a plurality of second features, wherein each of the second features indicates a prediction related to an acceptance status of the given product-related data structure by at least one model for a respective training period; and aggregating at least the second features to obtain a classification related to an aggregate acceptance status of the given product-related data structure. A weighting of at least some of the first and second features can be learned during a training phase.
1. A method, comprising:
obtaining data for a given product-related data structure;
evaluating a plurality of first features related to the given product-related data structure using the obtained data;
applying the plurality of first features related to the given product-related data structure to a plurality of models to obtain a corresponding plurality of second features, wherein each of the plurality of second features indicates a prediction related to an acceptance status of the given product-related data structure by a respective one of the models for a respective training period, wherein the plurality of models is trained using training data from a respective one of a plurality of different training periods, wherein each different training period comprises a different time duration of the training data, and wherein the plurality of first features is distinct from the respective time duration of the training data for the plurality of models, wherein the plurality of second features comprises respective ones of a plurality of acceptance status predictions, associated with respective ones of the plurality of different training periods, wherein the plurality of acceptance status predictions and one or more of the plurality of first features are applied to a classification engine that generates an aggregate acceptance status; and
aggregating at least the plurality of second features to obtain a classification related to the aggregate acceptance status of the given product-related data structure;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2. The method of claim 1 , wherein the acceptance status comprises one or more of an automatically accepted status, an automatically denied status and an additional review required status.
3. The method of claim 1 , wherein the aggregating comprises one or more of: (i) applying the plurality of second features to a machine learning model; (ii) determining an aggregate acceptance score based on an acceptance score assigned by each of the plurality of models trained using the training data from a respective one of the plurality of different training periods; (iii) determining whether a threshold number of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status; and (iv) determining whether a majority of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status.
4. The method of claim 1 , wherein the classification comprises one or more of a binary value, a probability value and a confidence value.
5. The method of claim 1 , wherein a training data set for the plurality of models comprises most recent training data for each of the respective ones of the plurality of different training periods.
6. The method of claim 1 , wherein a weighting of one or more of the first features and one or more of the second features is learned during a training phase.
7. The method of claim 1 , further comprising testing the plurality of models for each of the plurality of different training periods using a testing data set.
8. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to implement the following steps:
obtaining data for a given product-related data structure;
evaluating a plurality of first features related to the given product-related data structure using the obtained data;
applying the plurality of first features related to the given product-related data structure to a plurality of models to obtain a corresponding plurality of second features, wherein each of the plurality of second features indicates a prediction related to an acceptance status of the given product-related data structure by a respective one of the models for a respective training period, wherein the plurality of models is trained using training data from a respective one of a plurality of different training periods, wherein each different training period comprises a different time duration of the training data, and wherein the plurality of first features is distinct from the respective time duration of the training data for the plurality of models, wherein the plurality of second features comprises respective ones of a plurality of acceptance status predictions, associated with respective ones of the plurality of different training periods, wherein the plurality of acceptance status predictions and one or more of the plurality of first features are applied to a classification engine that generates an aggregate acceptance status; and
aggregating at least the plurality of second features to obtain a classification related to the aggregate acceptance status of the given product-related data structure.
9. The apparatus of claim 8 , wherein the acceptance status comprises one or more of an automatically accepted status, an automatically denied status and an additional review required status.
10. The apparatus of claim 8 , wherein the aggregating comprises one or more of: (i) applying the plurality of second features to a machine learning model; (ii) determining an aggregate acceptance score based on an acceptance score assigned by each of the plurality of models trained using the training data from a respective one of the plurality of different training periods; (iii) determining whether a threshold number of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status; and (iv) determining whether a majority of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status.
11. The apparatus of claim 8 , wherein the classification comprises one or more of a binary value, a probability value and a confidence value.
12. The apparatus of claim 8 , wherein a training data set for the plurality of models comprises most recent training data for each of the respective ones of the plurality of different training periods.
13. The apparatus of claim 8 , wherein a weighting of one or more of the first features and one or more of the second features is learned during a training phase.
14. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to performthe following steps:
obtaining data for a given product-related data structure;
evaluating a plurality of first features related to the given product-related data structure using the obtained data;
applying the plurality of first features related to the given product-related data structure to a plurality of models to obtain a corresponding plurality of second features, wherein each of the plurality of second features indicates a prediction related to an acceptance status of the given product-related data structure by a respective one of the models for a respective training period, wherein the plurality of models is trained using training data from a respective one of a plurality of different training periods, wherein each different training period comprises a different time duration of the training data, and wherein the plurality of first features is distinct from the respective time duration of the training data for the plurality of models, wherein the plurality of second features comprises respective ones of a plurality of acceptance status predictions, associated with respective ones of the plurality of different training periods, wherein the plurality of acceptance status predictions and one or more of the plurality of first features are applied to a classification engine that generates an aggregate acceptance status; and
aggregating at least the plurality of second features to obtain a classification related to the aggregate acceptance status of the given product-related data structure.
15. The non-transitory processor-readable storage medium of claim 14 , wherein the aggregating comprises one or more of: (i) applying the plurality of second features to a machine learning model; (ii) determining an aggregate acceptance score based on an acceptance score assigned by each of the plurality of models trained using the training data from a respective one of the plurality of different training periods; (iii) determining whether a threshold number of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status; and (iv) determining whether a majority of the plurality of models trained using the training data from a respective one of the plurality of different training periods had a given acceptance status.
16. The non-transitory processor-readable storage medium of claim 14 , wherein a training data set for the plurality of models comprises most recent training data for each of the respective ones of the plurality of different training periods.
17. The non-transitory processor-readable storage medium of claim 14 , wherein a weighting of one or more of the first features and one or more of the second features is learned during a training phase.
18. The method of claim 1 , wherein the plurality of first features related to the given product-related data structure comprises two or more of: at least one order feature, at least one product feature and at least one account feature.
19. The method of claim 1 , wherein the classification engine comprises at least one prediction model and wherein the plurality of second features are applied to the at least one prediction model.
20. The apparatus of claim 8 , wherein the classification engine comprises at least one prediction model and wherein the plurality of second features are applied to the at least one prediction model.