| CPC G06N 20/00 (2019.01) [G06F 16/2477 (2019.01); G06F 16/904 (2019.01)] | 19 Claims |

|
1. A computer-implemented method comprising:
identifying a black box time series model;
receiving training data for the black box time series model;
training the black box time series model using the training data received for the black box time series model;
predicting one or more time instances using the black box time series model, resulting in predicted data;
selecting a predicted time instance from the predicted data;
generating a set of white box time series models based on the black box time series model;
selecting a preferred white box time series model from the set of white box time series models based on a difference between each white box time series model and the black box time series model;
analyzing behavior of the preferred white box time series model; and
generating, based on the analyzed behavior and the training data for the black box time series model, an explanation illustrating why the black box time series model forecasted the predicted time instance, wherein generating the set of white box time series models comprises:
generating a set of model parameters for each of a plurality of types of white box time series models, resulting in a plurality of initial white box time series models; and
training each of the initial white box time series models using the training data for the black box time series model, the predicted data and the set of model parameters, resulting in a plurality of trained white box time series models, wherein generating the explanation for the black box time series model comprises:
determining a trend, seasonality, and residual of the black box time series model based on the analyzed behavior of the preferred white box time series model; and
generating a visualization of the black box time series model.
|