Churners Prediction Based on Mining the Content of Social Network Taxonomy
Asia Mahdi Naser Alzubaidi1, Eman Salih Al-Shamery2
1Asia Mahdi Naser Alzubaidi, Department of Computer Science, College of Science, Karbala University, Karbala, Iraq.
2Eman Salih Al-Shamery, Department of Information Technology Software, Babylon College of University, Iraq.
Manuscript received on 18 September 2019 | Revised Manuscript received on 05 October 2019 | Manuscript Published on 11 October 2019 | PP: 341-351 | Volume-8 Issue-2S10 September 2019 | Retrieval Number: B10560982S1019/2019©BEIESP | DOI: 10.35940/ijrte.B1056.0982S1019
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Churner Customer is a main tricky and one of the most important issues for large companies, due to the straight impact on the incomes of the companies especially in the telecom domain, companies are searching for advance strategies to predict churn/non-churn customer. This research focuses on the construction of a predictive model to identify each customer as churner or not and gain additional insights about their service consumers. The main contribution is to overcome the limitation of independently based on data mining strategies by developing approaches and derived network metrics such as centrality and connectivity between customers to incorporate network mining with traditional data mining. Social network measurements e.g. Leverage, flow Bet, Page Rank, Cluster Coefficients and Eccentricity are joined with other attributes in the original network dataset to enhance the performance of the proposed methodology. The risk of churn can be predictive by preparing an extensive cleaning the raw data for churn modeling, It divides customers into clusters based on Gower distance and k-medoids algorithm to help understand and predict churner users, classification model using Extreme Gradient Boosting “XGBoost”, assessment the model performance by computation the centralities metrics as new attributes appended to the original network dataset. Experiments conducted on Telecom shows that with an average value of all statistics accuracy not lower than 98.27%, while the average accuracy for the original dataset with it is clusters is not exceeded than 0.97%. The proposed method for churners detection which combines social impacts and network contents based on clustering significantly improved the prediction accuracy for telecom dataset as compared to prediction using the call log details, network information without implement of clustering , thus validate the hypothesis that combining social network attributes and Call/SMS information of the users for churn prediction could yields substantially improved of customer churn prediction. General Terms: Theory and Applications of Data Mining, Dimensionality Reduction, Business Analytics, Machine Learning, Supervised Statistical Learning.
Keywords: Churn Prediction, Mobile Social Network Analysis, Churn in Telecom, Social Network Analysis, eXtreme Gradient Boosting Algorithm (XGBoost), Centrality Metrics, Mobile Network.
Scope of the Article: Social Network