US 11,902,114 B2
System and method for predicting and reducing subscriber churn
Kamakshi Sridhar, Fremont, CA (US); Lars Anton Gunnarsson, Bangkok (TH); Alexander Havang, Malmo (SE); Pavle Mihajlovic, Waterloo (CA); and Kavi Kanasupramaniam, Waterloo (CA)
Filed by Sandvine Corporation, Waterloo (CA)
Filed on Dec. 8, 2021, as Appl. No. 17/545,162.
Application 17/545,162 is a continuation of application No. 16/598,112, filed on Oct. 10, 2019, granted, now 11,240,125.
Claims priority of provisional application 62/743,844, filed on Oct. 10, 2018.
Prior Publication US 2022/0131770 A1, Apr. 28, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. H04L 41/5067 (2022.01); H04L 41/147 (2022.01); H04L 41/5061 (2022.01); H04L 67/50 (2022.01)
CPC H04L 41/5067 (2013.01) [H04L 41/147 (2013.01); H04L 41/5064 (2013.01); H04L 67/535 (2022.05)] 20 Claims
OG exemplary drawing
 
1. A method for creating a model for predicting and reducing subscriber churn in a computer network, the method comprising:
for a predetermined time period:
retrieving traffic flow data per subscriber for a plurality of subscribers in the computer network;
determining at least one subscriber metric per subscriber from the traffic flow data;
determining whether the at least one subscriber metric includes data points to be condensed;
determining at least one systemic feature associated with the plurality of subscribers, wherein the at least one systemic feature is based on an access network used by the computer network; and
storing the at least one amalgamated metric and feature;
on reaching the predetermined time period create the model by:
analyzing the at least one subscriber metric and the at least one systemic feature for the predetermined time period and determining a relationship between the at least one subscriber metric, the subscriber quality of experience and the at least one systemic feature;
predicting, per subscriber, whether the subscriber is going to churn within a churn period in the future based on the analysis;
validating the prediction by determining whether the subscriber actually churned during the churn period; and
creating the model based on the validated predictions.