[Paper Review] A comparative study of social network classifiers for predicting churn in the telecommunication industry

María Óskarsdóttir, Cristián Bravo+4 more|arXiv (Cornell University)|Aug 18, 2016

Customer churn and segmentation41 references17 citations

TL;DR

This study evaluates relational learning methods for predicting customer churn in telecom using call detail record (CDR) data. It finds that collective inference does not improve performance, and the network-only link-based classifier—using logistic regression on link-based features—achieves the best results among tested relational classifiers.

ABSTRACT

Relational learning in networked data has been shown to be effective in a number of studies. Relational learners, composed of relational classifiers and collective inference methods, enable the inference of nodes in a network given the existence and strength of links to other nodes. These methods have been adapted to predict customer churn in telecommunication companies showing that incorporating them may give more accurate predictions. In this research, the performance of a variety of relational learners is compared by applying them to a number of CDR datasets originating from the telecommunication industry, with the goal to rank them as a whole and investigate the effects of relational classifiers and collective inference methods separately. Our results show that collective inference methods do not improve the performance of relational classifiers and the best performing relational classifier is the network-only link-based classifier, which builds a logistic model using link-based measures for the nodes in the network.

Motivation & Objective

• To evaluate the effectiveness of relational learning methods in predicting customer churn within telecom networks.
• To compare the performance of various relational classifiers and collective inference techniques on real-world CDR datasets.
• To determine whether collective inference enhances prediction accuracy beyond standalone relational classifiers.
• To identify the most effective relational learning approach for churn prediction in telecommunication contexts.

Proposed method

• The study applies multiple relational classifiers to call detail record (CDR) datasets from telecom providers.
• It evaluates link-based measures as features for logistic regression models within a network-only framework.
• Collective inference methods are applied to assess their impact on prediction performance.
• Performance is measured across multiple datasets to ensure robustness and generalizability.
• The analysis isolates the contributions of relational classifiers and collective inference to identify their individual effects.

Experimental results

Research questions

• RQ1 How do different relational classifiers perform in predicting customer churn using CDR data?
• RQ2 Does the integration of collective inference improve the accuracy of relational classifiers in churn prediction?
• RQ3 Which specific link-based features contribute most effectively to churn prediction models?
• RQ4 Is the network-only link-based classifier superior to models combining relational learning with collective inference?

Key findings

• Collective inference methods do not improve the performance of relational classifiers in predicting customer churn.
• The network-only link-based classifier, which uses logistic regression on link-based features, achieves the highest prediction accuracy.
• Relational classifiers that incorporate network structure outperform non-relational baselines in churn prediction tasks.
• The best-performing model relies solely on link-based measures without additional collective inference.