Predicting ledger revenue change behavior of clients receiving services

Patent metadata

Field	Value
Publication number	US-11062334-B2
Application number	US-201715614146-A
Country	US
Kind code	B2
Filing date	Jun 5, 2017
Priority date	Jun 5, 2017
Publication date	Jul 13, 2021
Grant date	Jul 13, 2021

Abstract

One embodiment provides a method for predicting revenue change in a ledger including receiving, by a processor device, revenue data with timestamps for a number of historical periods at a particular level, with attributes of the particular level and a percentage of the required revenue change. The data is filtered. The filtered data is aggregated at the particular level for a selected prediction. A sliding window of the number of historical periods is moved over business periods, creating a data point for each historical period temporal window by extracting features. A required target output is created for each data point for at least one future time period. A statistical classification model is trained to predict the revenue change. A set of recent histories is converted into a quantitative health value.

First claim

What is claimed is:

1. A method for machine learned prediction of future changes applicable to computing services for a plurality of accounts comprising:
receiving, by a processor device, data with timestamps for a number of historical periods at a particular level, with attributes of the particular level and a percentage of a required revenue change;
filtering, by the processing device, the data by removing invalid values for the attributes for creating filtered data;
aggregating, by the processing device, the filtered data at the particular level for a selected future prediction for generating aggregated data;
creating, by the processing device, a data point, from the aggregated data, for each historical period temporal window by extracting features based on moving a sliding window of the number of historical periods over business periods;
creating, by the processing device, a required target output for each data point for at least one future time period;
training, by the processing device, a statistical classification model by using:
machine learning processing that trains a plurality of boosted classification trees for learned gradient boosted classifiers that predict future changes applicable to computing services for a plurality of accounts, wherein the learned gradient boosted classifiers perform a search over a parameter that trades off between precision and recall during processing to obtain the trained statistical classification model that provides a maximum precision for a particular minimum recall based on focusing on areas of a parameter space, including the parameter, that have higher chances of attaining maximum objective value, and
a weighted loss function, employed over each data point, that returns a combination of losses where each loss of the combination of losses is separately weighted for each data point; and
for each of the data points: determining, by the processing device, learned outputs, from the trained statistical classification model, that include prediction and probability for each data point of predicted future changes applicable to computing services for the plurality of accounts metrics for predicted shrinking and abandoned accounts of for the plurality of accounts.

2. The method of claim 1 , wherein the required revenue change comprises one of a growth and a shrinkage at a particular percentage, and the parameter space includes parameters having higher probability for attaining a maximum objective value.

3. The method of claim 1 , wherein the selected future prediction comprises one of a prediction that a vendor will be abandoned within a next one or more periods unless action is taken by the vendor.

4. The method of claim 1 , wherein the required revenue change comprises a machine learned prediction at one of: an accounts or customer level, an offerings level, and an accounts-offerings level.

5. The method of claim 1 , wherein:
aggregating the filtered data at the particular level for the selected future prediction is performed at the required revenue change, at a required percentage, and at a required level;
the trained statistical classification model modifies the parameter with a modified parameterized objective function, such that the parameter balances tradeoff between precision and accuracy; and
the modified parameterized objective function is based on the learned gradient boosted classifiers predictions of false positives and false negatives.

6. The method of claim 1 , wherein creating the required target output is performed at the required revenue change and required percentage, and the weighted loss function provides that losses over data points that have particular values are provided more weight in minimization.

7. The method of claim 1 , wherein outputted values are binary predictions and a prediction probability for each data point in a test dataset are obtained using the trained statistical classification model.

8. The method of claim 7 , wherein one output comprises a ranked list in descending order of the prediction probability at the required level and provided to an electronic device.

9. The method of claim 1 , wherein features used to train the statistical classification model comprise revenues of three time periods used for the selected future prediction, a quotient of revenue in one period divided by an average revenue in a preceding two periods, a reverse quotient in one period divided by an average in a succeeding two periods, and a plurality of combinations and conversions comprising at least one of applying a log function, a sign of a quotient, a sign of a reverse quotient, and a number of positive quarters revenue out of three quarters used for the selected future prediction.

10. The method of claim 1 , wherein the selected future prediction is based on a specified condition that is predicted to hold sometime within the next two periods after the three periods on which the selected future prediction is based.

11. A computer program product for machine learned prediction of future changes applicable to computing services for a plurality of accounts, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
receive, by the processor, data with timestamps for a number of historical periods at a particular level, with attributes of the particular level and a percentage of a required revenue change;
filter, by the processor, the data by removing invalid values for the attributes for creating filtered data;
aggregate, by the processor, the filtered data at the particular level for a selected future prediction for generating aggregated data;
create, by the processor, a data point, from the aggregated data, for each historical period temporal window by extracting features based on moving a sliding window of the number of historical periods over business periods;
create, by the processor, a required target output for each data point for at least one future time period;
train, by the processor, a statistical classification model by using:
machine learning processing that trains a plurality of boosted classification trees for learned gradient boosted classifiers that predict future changes applicable to computing services for a plurality of accounts, wherein the learned gradient boosted classifiers perform a search over a parameter that trades off between precision and recall during processing to obtain the trained statistical classification model that provides a maximum precision for a particular minimum recall based on focusing on areas of a parameter space, including the parameter, that have higher chances of attaining maximum objective value, and
a weighted loss function, employed over each data point, that returns a combination of losses where each loss of the combination of losses is separately weighted for each data point; and
for each of the data points: determine, by the processor, learned outputs, from the trained statistical classification model, that include prediction and probability of predicted future changes applicable to computing services for the plurality of accounts metrics for predicted shrinking and abandoned accounts of for the plurality of accounts.

12. The computer program product of claim 11 , wherein:
the required revenue change comprises one of: a growth, a shrinkage at a particular percentage, and a machine learned prediction at one of: an accounts or customer level, an offerings level, and an accounts-offerings level; and
the parameter space includes parameters having higher probability for attaining a maximum objective value.

Assignees

• IBM

Inventors

• Blomberg Jeanette L
• Maurya Abhinav
• Megahed Aly
• Strong Jr Hovey R

Classifications

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06N3/126

  Evolutionary algorithms, e.g. genetic algorithms or genetic programming · CPC title

• G06Q10/04

  Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" (market predictions or forecasting for commercial activities G06Q30/0202) · CPC title

• G06Q40/12

  Accounting · CPC title