Please use this identifier to cite or link to this item:
Title: Churn prediction in telecommunication with social network analysis and MapReduce
Authors: Nguyen, Ngoc Tram Anh
Keywords: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
Issue Date: 2016
Abstract: In this day where telecommunication is getting saturated due to the same pricing model applied by most telcos, it is very easy for customers to leave one telco and join a competitive one. Churn prediction is a data mining technique to predict the probability of a customers wanting to leave the service. In this project, churn prediction classifier is implemented with data from an anonymous telecommunication company. The classifier is a binary classification with 2 labels churn and non-churn. We aggregate the data mining features from Call Detail Records (CDR) with basic features such as number of messages in a month, total duration of incoming/outgoing calls in a month, etc. Besides these basic features, graph theory features (Label Propagation and PageRank) are also incorporated in the feature selection method. With the huge amount of data, MapReduce is used to parallelize and partition graph computation such that graph size of 600000 nodes and more can be run comfortably in a personal computer. We achieve commendable results for the classification with all classifiers return around 90% accuracy and more. The classifiers used are Naïve Bayes, Logistic KNN, Logistic Regression, Decision Tree, Random Forest and Bagging. Logistic Regression consistently outperforms other classifiers with the highest result at 96.9% accuracy with AUC score of 0.988. We are confident that the telco will make profit in the long run if they offer these highly accurate potential churners attractive packages to keep them in the service.
Rights: Nanyang Technological University
Fulltext Permission: restricted
Fulltext Availability: With Fulltext
Appears in Collections:SCSE Student Reports (FYP/IA/PA/PI)

Files in This Item:
File Description SizeFormat 
  Restricted Access
3.81 MBAdobe PDFView/Open

Page view(s)

checked on Sep 22, 2020


checked on Sep 22, 2020

Google ScholarTM


Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.