US 11,941,496 B2
Providing predictions based on a prediction accuracy model using machine learning
Manish Anand Bhide, Hyderabad (IN); Venkata R Madugundu, Hyderabad (IN); Harivansh Kumar, Hyderabad (IN); and Prem Piyush Goyal, Hyderabad (IN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Mar. 19, 2020, as Appl. No. 16/823,477.
Prior Publication US 2021/0295204 A1, Sep. 23, 2021
Int. Cl. G06N 20/00 (2019.01); G06N 5/04 (2023.01)
CPC G06N 20/00 (2019.01) [G06N 5/04 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A computer-implemented method for machine-learning model accuracy, comprising:
determining, by a computer processor and using a prediction accuracy machine-learning model, a first confidence that a first prediction generated, for a first data point of a client transaction, and by a classifier of a machine-learning model is accurate by:
generating, by the computer processor, prediction training data comprising, for each of a plurality of training transactions of the machine-learning model:
a data point;
a training prediction made by the classifier for the data point;
an indication whether the training prediction is accurate;
a first probability that the machine learning model determines that the training prediction is accurate; and
a second probability that that the machine learning model determines that an alternative prediction is accurate;
training, by the computer processor, the prediction accuracy machine-learning model to determine a client confidence that a client prediction generated by the machine-learning model is accurate using the prediction training data; and
generating, using the prediction accuracy machine-learning model:
a second prediction for the first data point; and
the first confidence, wherein the first confidence is equal to a second confidence that the second prediction is accurate;
wherein the first prediction represents a machine-learning classification of a corresponding client transaction.