Interpretable machine learning with an ensemble of gradient boosting machines

A method for the local and global interpretation of a black-box model on basis well-known generalized additive models is proposed. It can be viewed as an extension or modification algorithm using neural model. The based ensemble gradient boosting machines (GBMs) such that each GBM learned single feature produces shape function feature. composed weighted sum separate GBMs resulting functions which form are built in parallel randomized decision trees depth 1, provide very simple architecture. Weights well features computed iteration by Lasso then updated means specific smoothing procedure. In contrast to model, provides weights explicit form, it simply trained. lot numerical experiments with implementing proposed synthetic real datasets demonstrate its efficiency properties interpretation.