ENBIS-17 in Naples

9 – 14 September 2017; Naples (Italy) Abstract submission: 21 November 2016 – 10 May 2017

Clustering Time Series from Call Networks to Predict Churn

13 September 2017, 10:30 – 10:50


Submitted by
María Óskarsdóttir
María Óskarsdóttir (KU Leuven), Tine Van Calster (KU Leuven), Bart Baesens (KU Leuven), Wilfried Lemahieu (KU Leuven), Jan Vanthienen (KU Leuven)
Accurately predicting potential churners is important in fast-moving and saturated markets such as the telecommunication industry. Being able to identify certain behavioral patterns that lead to churn, is equally important, because it allows the organization to make arrangements for retention in a timely manner. Moreover, previous research has shown that the decision to leave one operator for another, is often influenced by the customer’s social circle. Therefore, features representing the churn status of their connections are usually good predictors of churn when it is treated as a binary classification problem, which is the traditional approach.
We propose a method to discover common behavior among churn-prone telecom customers and to distinguish them from loyal customers. More precisely, we use call detail records (CDR) of the customers of a telecommunication provider to build call networks on a weekly basis over the period of six months. From each network, we extract features based on each customer’s connections within the network, resulting in individual time series of link-based measures. In order to identify common behavior, we then apply time series clustering techniques and assign the customers to different groups at subsequent time points. Finally, we analyze how the customers move between the clusters to identify frequent patterns, especially amongst churners.
Our approach offers the possibility to discover behavioral patterns of potential churners, depending on the temporal aspect of phone usage as well as individual call networks. The result, once the patterns have been extracted, is a model that is simple in deployment and easily expandable.

Return to programme