Mining Telecommunication Networks to Enhance Customer Lifetime Predictions

Abstract

Customer retention has become a necessity in many markets, including mobile telecommunications. As it becomes easier for customers to switch providers, the providers seek to improve prediction models in an effort to intervene with potential churners. Many studies have evaluated different models seeking any improvement to prediction accuracy. This study proposes that the attributes, not the model, need to be reconsidered. By representing call detail records as a social network of customers, network attributes can be extracted for use in various traditional prediction models. The use of network attributes exhibits a significant increase in the area under the receiver operating curve (AUC) when compared to using just individual customer attributes.

APATE: A novel approach for automated credit card transaction fraud detection using network-based extensions

“…prediction (Verbeke et al, 2011) and survival analysis (Backiel et al, 2014). Nevertheless, fraud detection in general is an atypical prediction task which requires a tailored approach to address and predict future fraud.…”

Social network analytics for churn prediction in telco: Model building, evaluation and network architecture

“…It is used on its own (Backiel et al, 2015;Dasgupta et al, 2008) and to produce scores that are then used as variables in non-relational classifiers, such as logistic regression and decision trees (Kim et al, 2014;Kusuma et al, 2013). Other works enrich their datasets with network measures for non-relational classifiers, instead of exploiting relational learners (Backiel et al, 2014;Benedek et al, 2014;Modani et al, 2013;Zhang et al, 2012). Verbeke et al (2014) For most of the papers in Table 2, SNA relies on a single network which is defined and constructed only once for the situation at hand.…”

“…In classical customer churn prediction (CCP) modeling, a binary classifier is applied to available customer data at the company to build a predictive model which assigns each customer a score representing their propensity of churning. Social network analytics (SNA) has become a substantial addition to this field, as studies show that, when the customer datasets contain network features in addition to customer attributes, the performance of CCP models is enhanced (Backiel et al, 2014;Kusuma et al, 2013;Richter et al, 2010). The network features are extracted from call networks and encapsulate both calling behavior and interactions between customers.…”

A comparative study of social network classifiers for predicting churn in the telecommunication industry

“…In recent studies where social effects have been incorporated in the models it has been shown that the model performance improves greatly. In some studies the datasets have been enriched with network variables, such as degree, transitivity and centrality, before building a model using binary classifiers [13], [14], [15], [16]. Including these kinds of variables in the dataset adds valuable information that is different from the local variables, and thus better models are produced.…”

“…Based on previous studies [15], [4] and expert knowledge, churn was defined as being inactive for 30 days and then the churn day was defined as the day on which the customer became inactive. This also ensures the consistency between datasets, since actual churn dates were only available for a couple of them.…”