System and method of cyclic boosting for explainable supervised machine learning

Patent metadata

Field	Value
Publication number	US-11544724-B1
Application number	US-201916653642-A
Country	US
Kind code	B1
Filing date	Oct 15, 2019
Priority date	Jan 9, 2019
Publication date	Jan 3, 2023
Grant date	Jan 3, 2023

Abstract

A system and method are disclosed including a computer and a processor and memory. The computer receives historical sales data comprising aggregated sales data for one or more items from one or more store for at least one past time period. The computer further trains a cyclic boosting model to learn model parameters by iteratively calculating for each feature and each bin factors for at least one full feature cycle. The computer further predicts one or more demand quantities during a prediction period by applying a prediction model to historical supply chain data, wherein a training period is earlier than the prediction period, and each of the one or more demand quantities is associated with at least one item of the one or more items and at least one stocking location of the one or more stocking locations during the prediction period and rendering a demand prediction feature explanation visualization.

First claim

What is claimed is:

1. A computer-implemented method of explainable supervised machine learning cyclic boosting for prediction and explanation of future customer demand quantities, comprising:
receiving, by a server comprising a processor and a memory, historical sales data comprising aggregated sales data for one or more items from one or more stores for at least one past time period, wherein the one or more stores maintain an inventory of the one or more items at one or more stocking locations;
binning categorical features from the historical sales data according to feature categories;
binning continuous features from the historical sales data;
training a cyclic boosting model to learn model parameters by iteratively calculating, for each feature and each bin, one or more factors for at least one full feature cycle,
wherein a training period is earlier than a prediction period, and each of one or more demand quantities is associated with at least one item of the one or more items and at least one stocking location of the one or more stocking locations during the prediction period, and
wherein training the cyclic boosting model to learn the model parameters by iteratively calculating, for each feature and each bin, the one or more factors for the at least one full feature cycle, comprises:
calculating partial factors g and corresponding aggregated factors f according to:

g

j
,
t

k

=

∑

x

j
,
i

∈

b
j
k

y
i

∑

x

j
,
i

∈

b
j
k

y
^

i
,
τ

;

 and

f

j
,
t

k

=

∏

s
=
1

t

g

j
,
s

k

,

wherein ŷ j is a predicted value of a target variable for each feature j and each bin k denoted by b j k of an observation x j,i , index t is a current iteration, index τ is a preceding iteration, and iterations of both the index t and the index τ are full feature cycles;
predicting the one or more demand quantities during the prediction period by training a prediction model on historical supply chain data; and
rendering, for display on a user interface, a demand prediction feature explanation visualization comprising a predicted demand and one or more features identified during the training of the prediction model that influence the predicted demand.

2. The computer-implemented method of claim 1 , further comprising:
rendering, for display on the user interface, one or more interactive graphical elements for selection of the one or more items and stores;
in response to the selection of the one or more items and stores, retrieving the one or more factors influencing the predicted demand; and
rendering, for display on the user interface, a visualization comprising one or more graphical elements identifying the one or more features and the one or more retrieved factors.

3. The computer-implemented method of claim 1 , further comprising:
rendering, for display on the user interface, one or more interactive graphical elements that provide for modifying one or more future states of the one or more features identified by the cyclic boosting model during the training; and
in response to modifying the one or more future states of the one or more features, modifying input values to represent a future scenario corresponding to the modified one or more future states of the one or more features.

4. The computer-implemented method of claim 1 , further comprising:
binning continuous features, wherein each bin has one or more of a same width or a same quantity of observations.

5. The computer-implemented method of claim 1 , further comprising:
calculating, for each of the iterations, from the partial factor g and an aggregated factor f t-1 , the predicted value ŷ i of the target variable for the current iteration according to:

y
^

i

=

μ
·

∏

j
=
1

p

f
j
k

⁢

with
⁢

k

=

{

x

j
,
i

∈

b
j
k

}

Assignees

• Jda Software Group Inc
• Blue Yonder Group Inc

Inventors

• Wick Felix Christopher
• Feindt Michael

Classifications

• G06Q30/0202 Primary

  Market predictions or forecasting for commercial activities · CPC title

• G06F3/0484

  for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range · CPC title

• G06F3/0482

  Interaction with lists of selectable items, e.g. menus · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N20/20

  Ensemble learning · CPC title