| CPC G06N 7/01 (2023.01) [G06N 5/022 (2013.01)] | 18 Claims |
|
1. A method, comprising:
processing nonlinear input data associated with an electronic data transaction with an ensemble of tree-based nonlinear machine learning models to generate an output at each leaf node of each tree-based nonlinear machine learning model, wherein the output is based on a traversal path of each tree-based nonlinear machine learning model in the ensemble of tree-based nonlinear machine learning models;
generating a high-dimensional embedding based on the output of each leaf node of each tree-based nonlinear machine learning model in the ensemble of tree-based nonlinear machine learning models, wherein the high-dimensional embedding encodes one or more nonlinear features associated with the nonlinear input data traversed in the traversal path of each tree-based nonlinear machine learning model in the ensemble of tree-based nonlinear machine learning models;
projecting the high-dimensional embedding into a lower-dimensional embedding by applying a dimensionality reduction function, wherein:
the dimensionality reduction function is based on a principal component analysis, and
the lower-dimensional embedding comprises a lower-dimensional representation of the one or more nonlinear features;
processing the lower-dimensional embedding with a Bayesian logistic regression machine learning model to generate a binary class prediction associated with the nonlinear input data;
determining a confidence for the binary class prediction with the Bayesian logistic regression machine learning model, wherein the confidence for the binary class prediction is based on a credible interval of the binary class prediction and the nonlinear input data; outputting:
the binary class prediction if the confidence is greater than or equal to a threshold; or
a flipped binary class prediction if the confidence is lower than the threshold; and
authorizing the electronic data transaction based on the binary class prediction or the flipped binary class prediction.
|