Churn Prediction in Telecom Industry using Social Network Analysis

DOI : 10.17577/IJERTCONV8IS15027

Download Full-Text PDF Cite this Publication

Text Only Version

Churn Prediction in Telecom Industry using Social Network Analysis

Mr. Varun E Research Scholar Department of CS &E AIT, Chikmagalur

Dr. PushpaRavikumar

Professor & Head Department of CS &E AIT, Chikmagalur

Abstract – Applied In the telecom portion, an enormous volume of data is being created each day in light of an immense client base. The telecom organizations consider that achieving new client is costlier than holding the current ones. This paper proposes an interpersonal organization examination based model used to distinguish the constructive key clients. The proposed approach assembles a telecom interpersonal organization; it investigates essential system includes, and distinguishes the network structures present in the system. The most significant piece of base mining is inner circle structures which share social structure, use structures and regular interests. The inner circle is a gathering of clients in the telecom interpersonal organization, they share comparative properties. The paper recognizes positive key clients, its club subgroup and that foundation is utilized to distinguish the churn in telecom client's system.

Keywords: Telecom social network, substructures, cliques.


    In the current world, a monstrous volume of data is being created by telecom associations at a really fast rate. There is an extent of telecom master fighting in the market to assemble their client share. The clients have various options as better and progressively moderate administrations. A complete goal of telecom associations is to support their clients and stay alive in an engaged business showcase. A client stir happens when an enormous degree of clients are not happy with the administration gave by telecom association.

    Telecom associations consider arranging move when the amount of customers administration under a particular level which may achieve a huge loss of salary. Client stir distinguishing proof is basic in the telecom division as telecom managers need to hold their productive clients and improve their Client Relationship. The most significant factor is holding existing clients, since procuring new client is as yet costlier than holding existing [1]. The telecom organizations ought to have associations model that perceive and hold their clients by offering moderate and great types of assistance [2].

    The informal community examination approach is utilized for agitate forecast in associations. Informal organization examination is the planning, estimating the connections and distinguishing the flood of correspondence between clients. The hubs present in the system are clients and the connection speaks to the connection between every clients. The informal organization investigation (SNA) approach gives scientific and visual methods to point by point examination. To get arrange and their member clients, the diverse centrality measures is utilized. The various estimates utilized for understanding the SNA properties are degree centrality, betweenness centrality, closeness centrality and Eigen vector centrality. The center centrality of informal community investigation is design present in organize connections, the impact on singular client, imperatives and practices.

    The regular interests of basic investigation is in the "sub-structures" that is available in a system. A significant number of the methodologies for understanding the structure of a system focus on how thick associations are available and stretched out to create bigger factions or sub-groupings. The inner circles are a calculation created to recognize how bigger structures are exacerbated from littler structures. The particular definition to coteries is on the off chance that there are n clients; at that point there exist every single imaginable tie among themselves and nearby one another.


    1. Churn prediction

      Client beat is a noteworthy issue and one of the most huge concerns for enormous associations [3]. Due to the quick effect on the livelihoods of the associations, especially in the telecom field, associations are attempting to make expects to anticipate possible clients for maintenance [4]. Thusly, finding factors that impact clients to agitate is basic to take basic exercises to diminish the beat. The work on client agitate expectation focuses on

      building a forecast model dependent on utilization conduct, use example and AI draws near. The a large portion of SNA model assembled so far focused on single social and homogeneous structure [5]. Another ordinarily utilized methodology for client agitate forecast is choice tree; there is an imperative in choice tree that it isn't reasonable for complex nonlinear associations between the qualities. The choice tree approach has preferences like it tends to be handily envisioned, can process clear cut and numerical information without earlier suspicions. A neural system based methodology is for forecast of agitate additionally can be created [6].

    2. Substructure mining

    The interpersonal organization contains numerous connections and hubs. The connections and hubs are consolidated with certain relations that can be demonstrated as a diagram. For the network mining these hubs and connections can give distinctive related networks. In interpersonal organization investigation (SNA), the foundation emulating is regularly viewed as a durable sub- network, for example, the factions, n-inner circles, n- groups, n-plexes, just as the semi clubs. The most existing methodology thinks about that there is just a single sort of connection in the system and the aftereffects of mining are autonomous of client inclinations. Anyway in actuality there are numerous heterogeneous social associations every relationship assume particular jobs. The relapse based methodology is utilized to discover the networks; however the methodology isn't appropriate for shrouded network structures. The occasion based methodologies are likewise proposed to identify the network; however they are not reasonable for telecom beat expectation network approach.


    Figure 1: Sample telecom data set

    This segment speaks to proposed agitate expectation approach. Figure 2 shows the model and depicts the means. The initial step is information pre- preparing, in information pre-handling covering approach is utilized to perform include choice [7]. The irregular timberland, boruta calculation, stepwise forward choice and stepwise in reverse choice calculations are utilized to separate significant highlights [8][9]. After element choice the comparability measures is determined, which yields nearness lattice. The subsequent stage is to assemble telecom informal organization and estimating diverse centrality esteems to distinguish the positive key client and negative key client. The last advance is to recognize the stir network. The informational collection of 24588 clients is appeared in the figure 1. The dataset is organized information which contains 76 distinct characteristics and 24588 records of clients.

    Identification of Most Suitable Data

    Identification of Most Suitable Data

    Data Semantics

    Data Semantics

    Feature Selection

    Feature Selection

    Development of Predictive Model

    Development of Predictive Model

    Validation of results

    Validation of results

    Figure 2: Churn prediction framework using social network analysis


    1. Construction of directed multi-relational heterogeneous network

      Telecom informal community speaks to various connections, communications among clients. The system developed for investigation object is multi-social and heterogeneous that is clients of different sort of administrations like post-paid and paid ahead of time. The telecom arrange built utilizing numerous characteristic

      from call detail record and the client data. The trait considered arecustomer_age, number_of_SMS in seven days, number_of_calls in seven days, duration_of_the_calls. Utilizing the eucledian separation comparability measure, the nearness grid is developed. The qualities present in the nearness framework speak to relationship separation between the clients in the system. The telecom interpersonal organization developed by utilizing all the properties is appeared in figure 3.

      Figure 3: Customers telecom social network

    2. Centrality measures

      Centrality measure is the most noteworthy undertaking in a diagram. It is utilized to locate the most significant hubs in the client's system. Distinctive centrality estimates utilized are degree, betweenness, and closeness and Eigen centrality. The centrality measure can be utilized to recognize positive key players and negative key players in the system.

      Given a telecom interpersonal organization with n clients, to such an extent that, adverse key client implies expelling that n-p set will bring about the system with least attachment. The key client positive methods it is maximally associated with all the hubs. The negative key players of size k=1 is 76899 having less centrality measures.

    3. Identifying subgroups

    The inner circle in a telecom interpersonal organization is a maximally complete foundation, the clubs with more noteworthy than size k is recognized as a subgroups. The figure 4 shows number of inner circles distinguished in telecom informal community, there are

    364 clubs or sub bunches in the given telecom interpersonal organization. The biggest coterie in the system is made out of 6 clients, they share basic properties. The various clubs of size k=3 to k=6 were found.

    Figure 4: Clique sub-structures present in the telecom social network

    There are 364 coteries of size k = 3 to 3 of size k = 6 are found in the telecom informal organization. The gathering of clients like 3196, 36928, 43962, 52389, 112066 and 185349 offer basic properties and all are near one another. Think about coterie with number of clients or hub n, to such an extent that no hub with more n-1 vertices found.

    Table 1: Number of cliques found in the overall network

    Clique size

    Number of cliques found









    Table 1 records all out number of factions found for the telecom informal community. There are 364 coteries of size k = 3 in the clients organize, 228 factions of size k = 4 is found. At the point when the quantity of client hubs expands, the size of longer factions will diminish.


This paper utilizes interpersonal organization investigation way to deal with discover bases. Clients are allotted with the centrality scores, which distinguishes positive key clients present in the system. In the wake of allotting scores the bases like factions are distinguished. In the event that the inner circle has positive key clients, at that point that faction is recognized as likely gathering and plausibility of stir suggestion is more in that bases. The outcomes contrasted and class mark adjusts 80% of churners present in that bases. The subgroups inner circles

distinguished follows exacting arrangement of gatherings, it tends to be reached out to increasingly loosen up n- cliques.


  1. Preeti K. Dalvi, Siddhi K. Khandge, AshishDeomore, AdityaBankar,

    V. A. Kanade, Analysis of customer churn prediction in telecom industry using decision trees and logistic regression, in Proc. IEEE Symposium on Colossal Data Analysis and Networking (CDAN), pp: 1 4, 2016.

  2. Hui Li, Deliang Yang, Lingling Yang, YaoLu, Xiaola Lin, Supervised Massive Data Analysis for Telecommunication Customer Churn Prediction, in Proc. IEEE International Conferences on Big Data and Cloud Computing, Social Computing and Networking, Sustainable Computing and Communications, pp: 163 169, 2016.

  3. Ammar A Ahmed, Dr. D. Maheswari linen, A Review And Analysis Of Churn Prediction Methods For Customer Retention In Telecom Industries, in Proc. IEEE International Conference on Advanced Computing and Communication Systems (ICACCS – 2017), January, pp: 06 07, 2017, Coimbatore, India.

  4. SepidehHassankhaniDolatabadi ,FarshidKeynia Designing of Customer and Employee Churn Prediction Model Based on Data Mining Method and Neural Predictor, The 2nd International Conference on Computer and Communication Systems, 2017.

  5. Dr. PushpaAnd Dr. Shobha Social Network Classifier for Churn Prediction in Telecom Data 2013 International Conference on Advanced Computing and Communication Systems (ICACCS – 2013), Dec. 19 21, 2013, Coimbatore, INDIA [978-1-4799-3506- 2/13 ©2013 IEEE].

  6. B.N.KrishnaSai ; T. Sasikala, "Predictive Analysis and Modeling of Customer Churn in Telecom using Machine Learning Technique, ICOEI 2019.

  7. Varun E, PushpaRavikumar,Chandana S, Spandana K M, An Efficient Technique for Feature Selection to Predict Customer Churn in telecom industry, ICAIT 2019.

  8. Varun E and PushpaRavikumar, Attribute Selection for Telecommunication Churn Prediction, International Journal of Engineering & Technology, 2018.

  9. S. Jagannatha, M. Niranjanamurthy, and P. Dayananda, Algorithm Approach: Modelling and Performance Analysis of Software System , Journal of Computational and Theoretical Nanoscience (American Scientific publishers), December 2018, Volume 15, Issue 15, PP. 33893397

  10. Varun E and PushpaRavikumar, Telecommunication Community Detection by Decomposing Network into n-Cliques, IEEE International Conference on Emerging Technological Trends (ICETT), 2016.

Leave a Reply