Peng Jiang

US Patent:

20200097439, Mar 26, 2020

Filed:

Sep 20, 2018

Appl. No.:

16/137200

Inventors:

- Eden Prairie MN, US
Damien Benveniste - Torrance CA, US
Peng Jiang - Torrance CA, US
Philip Remmele - El Segundo CA, US
Junqing Wu - Playa Del Rey CA, US
Mike Zhang - Fullerton CA, US

International Classification:

G06F 15/18
G06F 17/11
G06K 9/62

Abstract:

Embodiments herein provide for a machine learning algorithm that generates models that are more interpretable and transparent than existing machine learning approaches. These embodiments identify, at a record level, the effect of individual input variables on the machine learning model. To provide those improvements, a reason code generator assigns monotonic relationships to a series of input variables, which are then incorporated into the machine learning algorithm as metadata. In some embodiments, the reason code generator creates records based on the monotonic relationships, which are used by the machine learning algorithm to generate predicted values. The reason code generator compares an original predicted value from the machine learning model to the predicted values from the machine learning model.