| CPC G06N 20/10 (2019.01) [G06F 16/2379 (2019.01); G06N 5/045 (2013.01)] | 20 Claims |
|
1. A method, comprising:
receiving, by a device, user information, wherein the user information includes data associated with an entity;
determining, by the device and based on a prediction model, a prediction output of an analysis of the user information,
wherein the prediction output indicates a probability associated with the entity failing to satisfy a set of criteria;
generating, by the device and based on a generator model, data representative of counterfactuals of parameters of the user information;
utilizing, by the device and based on the generator model, vector conversion to convert the data representative of the counterfactuals of the parameters of the user information into a plurality of counterfactual explanations associated with the prediction output and the user information,
wherein each counterfactual explanation, of the plurality of counterfactual explanations, describes a respective scenario that would enable the entity to satisfy the set of criteria, and
wherein the generator model is trained based on a plurality of labeled counterfactuals associated with historical outputs of the analysis;
clustering, by the device and according to a clustering model, the plurality of counterfactual explanations into clusters of counterfactual explanations;
selecting, by the device and based on a classification model, one or more counterfactual explanations from a cluster of the clusters of counterfactual explanations based on the prediction output and a relevance score of the selected one or more counterfactual explanations,
wherein the relevance score is determined based on the user information and a confidence score associated with the clustering of the plurality of counterfactual explanations and a confidence score associated with the prediction model;
providing, by the device, a request for feedback associated with the selected one or more counterfactual explanation;
receiving, by the device, feedback data associated with the request for feedback;
updating, by the device, a data structure associated with the clustering model based on the feedback data and the counterfactual explanation to form an updated labeled counterfactual explanation data structure,
populating, by the device, the updated labeled counterfactual explanation data structure with at least a threshold quantity and a threshold percentage of counterfactual explanations associated with historical outputs of the analysis,
training, by the device, a machine learning model to generate a trained machine learning model based on the updated labeled counterfactual explanation data structure including a desired quantity of labeled counterfactual explanations,
generating, by the device, using the trained machine learning model, optimal counterfactual explanations for subsequent prediction outputs of the prediction model based on the updated labeled counterfactual explanation data structure; and
performing, by the device, one or more actions associated with the updated labeled counterfactual explanation data structure based on the optimal counterfactual explanations.
|