Back to results
US 11,232,465 B2Granted
Churn prediction with machine learning
AssigneeAirship Group, Inc.
InventorsElizabeth Marjory Orr, Gary Todd Johnson, Neel Banerjee
Life of this patent
Estimating the term…
Abstract
Disclosed is a churn prediction system that predicts with a high level of accuracy which users will and which users will not stop opening the app over a 30-day time period. To this end a model is created using historical event data where the churn-related behavior of each user is known. New event data is then applied to the model to determine the likelihood of each user churning in the future. With these prediction scores a user is then qualified as falling into one of three classifications: low-risk, medium-risk, or high-risk of churn.
Claims20
A computer implemented method, comprising:
- receiving a device identifier and an event descriptor, wherein the device identifier is associated with a target device, and wherein the event descriptor describes an event associated with an application that occurred for the target device;
- identifying historical event data describing events associated with the application that have occurred for a plurality of user devices by a cutoff date, the events including one or more open events and send events;
- using the historical event data to train the machine learning model to predict a likelihood of users of the application churning after the cutoff date, the training of the machine learning model comprising, for each particular user device of the plurality of user devices:
- determining a shifted cutoff date for the particular user device, the shifted cutoff date being a predetermined time before the cutoff date,
- determining a first time period of historical events for the particular user device ending at the shifted cutoff date;
- determining a second time period of historical events for the particular user device extending from the shifted cutoff date to the cutoff date;
- identifying first event data describing events occurring for the particular user device within the first time period;
- assigning the first event data of the particular user device to a positive or negative training set using second event data describing user events occurring for the particular user device within the second time period, wherein the assigning comprises:
- responsive to the second activity data indicating the particular user device churned during the second time period, assigning the first activity data to the positive training set;
- and
- responsive to the second activity data indicating the particular user device did not churn during the second time period, assigning the first activity data to the negative training set;
- training the machine learning model based on the positive training set and the negative training set generated from the historical event data;
- applying, using the event descriptor, the trained machine learning model for determining the likelihood of the target device churning within a future time period after the cutoff date;
- and
- transmitting the determined likelihood of the target device churning.
12 more elements
A system comprising:
- a processor;
- and
- a memory comprising stored instructions, the instructions executable by the processor for causing the processor to:
- receive a user identifier and an event descriptor, wherein the user identifier uniquely identifies a target user associated with a target device, and wherein the event descriptor describes an event associated with an application that occurred for the target device;
- identify historical event data describing events associated with the application that have occurred for a plurality of user devices by a cutoff date, the events including one or more open events and send events
- use the historical event data to train the machine learning model to predict a likelihood of users of the application churning after the cutoff date the training of the machine learning model comprising, for each particular user device of the plurality of user devices:
- determine a shifted cutoff for the particular user device, the shifted cutoff date being a predetermined time before the cutoff date,
- determine a first time period of historical events for the particular user device ending at the shifted cutoff date;
- determine a second time period of historical events for the particular user device extending from the shifted cutoff date to the cutoff date;
- identify first event data describing events occurring for the particular user device within the first time period;
- assign the first event data of the particular user device to a positive or negative training set using second event data describing user events occurring for the particular user device within the second time period, wherein the assigning comprises:
- responsive to the second activity data indicating the particular user device churned during the second time period, assign the first activity data to the positive training set;
- and
- responsive to the second activity data indicating the particular user device did not churn during the second time period, assign the first activity data to the negative training set;
- train the machine learning model based on the positive training set and the negative training set generated from the historical event data;
- apply, using the event descriptor, the trained machine learning model for determining the likelihood of the target user churning within a future time period;
- and
- transmit the likelihood of the target user churning.
15 more elements
A computer program product embodied in a non-transitory computer readable storage medium and comprising computer instructions that when executed by a processor causes the processor to:
- receive a user identifier and an event descriptor, wherein the user identifier uniquely identifies a target user associated with a target device, and wherein the event descriptor describes an event associated with an application that occurred for the target device;
- identify historical event data describing events associated with the application that have occurred for a plurality of user devices by a cutoff date, the events including one or more open events and send events
- use the historical event to train the machine learning model to predict a likelihood of users of the application churning after the cutoff date, the training of the machine learning model comprising, for each particular user device of the plurality of user devices:
- determine a shifted cutoff date for the particular user device, the shifted cutoff date being a predetermined time before the cutoff date,
- determine a first time period of historical events for the particular user device ending at the shifted cutoff date;
- determine a second time period of historical events for the particular user device extending from the shifted cutoff date to the cutoff date;
- identify first event data describing events occurring for the particular user device within the first time period;
- assign the first event data of the particular user device to a positive or negative training set using second event data describing user events occurring for the particular user device within the second time period, wherein the assigning comprises:
- responsive to the second activity data indicating the particular user device churned during the second time period, assign the first activity data to the positive training set;
- and
- responsive to the second activity data indicating the particular user device did not churn during the second time period, assign the first activity data to the negative training set;
- train the machine learning model based on the positive training set and the negative training set generated from the historical event data;
- apply, using the event descriptor, the trained machine learning model for determining the likelihood of the target user churning within a future time period;
- and
- transmit the likelihood of the target user churning.
12 more elements
Dependencies are inferred from the claim text.