Telco Churn Prediction with Big Data

Abstract

We show that telco big data can make churn prediction much more easier from the 3V's perspectives: Volume, Variety, Velocity. Experimental results confirm that the prediction performance has been significantly improved by using a large volume of training data, a large variety of features from both business support systems (BSS) and operations support systems (OSS), and a high velocity of processing new coming data. We have deployed this churn prediction system in one of the biggest mobile operators in China. From millions of active customers, this system can provide a list of prepaid customers who are most likely to churn in the next month, having 0.96 precision for the top 50000 predicted churners in the list. Automatic matching retention campaigns with the targeted potential churners significantly boost their recharge rates, leading to a big business value.

Differential privacy: its technological prescriptive using big data

“…The first, second and third generation of privacy protection techniques may be explored and compared in telco big data platform. First generation of privacy protection technique removes or replaces the explicitly sensitive identifiers of customers, second generation publishes a sanitized database with certain anonymity or diversity requirements and the third generation is DP [12,[32][33][34]. 4.…”

A fuzzy based churn prediction and retention model for prepaid customers in telecom industry

“…Feature selection methods have been adopted in past by different researchers to select most important features for accurate churn prediction model building. For prediction model building, 84 different features have been selected in [18], however the dataset under experiments contains various features other than CDRs. In this work, supervised, semi-supervised and unsupervised learning algorithms have been used to get most important features.…”

“…In order to combat customer churn, a number of Machine Learning (ML) based churn prediction models have been proposed in the recent past. To name a few, Multilayer Perceptron [4][5][6], Linear Regression (LR) [7], classification based on Support Vector Machine (SVM) [8][9][10], Association Rules [11], advanced rule induction [12], Decision Tree (DT) [13], ensemble of hybrid methods [14], churn prediction by feature selection techniques [15,16], Bayesian network classifiers [17] and improved balanced random forests [18]. The primary objective of these previously reported prediction models is to utilize large amount of telecom data to identify potential churners.…”

“…Firstly, the feature selection methods adopted in majority of the previous models neglected, with the exception of [13], the information rich variables present in call details record (CDRs) for model development. Instead, features from service logs [14], complaints data [11], customer demographics [12], bill and payment data [15], contractual data [13] and operation support system (OSS) data [18] has been used. Moreover, the selection of important features in the past was done only through statistical methods [11].…”

Survey of Data Mining Techniques Used for Real Time Churn Prediction

“…Following the previous paragraphs line, [31] many classifiers have been adopted for churn prediction, including logistic regression, decision trees, boosting algorithms (e.g., variants of adaboost), boosted trees (gradient boosted decision trees) or random forest, neural networks, evolutionary computation (e.g., genetic algorithm and ant colony optimization), ensemble of support vector machines, and ensemble of hybrid methods. Most customer behavior features are extracted from BSS, including call detailed records (call number, start time, data usage, etc.…”