Content uploaded by Niyamaddin Taghiyev
Author content
All content in this area was uploaded by Niyamaddin Taghiyev on Jun 02, 2021
Content may be subject to copyright.
N.R.TAGHIYEV
CHURN INDICATORS DESIGN IN TELECOM
taghiyevniyam@gmail.com
Abstract. In the telecom industry, retaining more customers than acquiring a new customer
has become one of the main priorities. Customers are treated more sensitively, especially as
there is more competition in this area. Therefore, they are forced to build and research a
customer churn prediction using various databases and algorithms to understand customers.
With this in mind, we have tried to take the most effective indicators from the possible source
data that will affect customers. This paper describes the process of developing derivative
indicators to determine the initial indicators and use them in the model of predict potential
churn in telecom companies. The main purpose of this research is to determine the design of
optimal churn indicators.
Keyword: Churn prediction, Churn indicators, Telecommunication industry, OSS, BSS,
Customer complaint
Introduction
The telecom industry is one of the most profitable companies today and
has the largest number of customers in the business. However, in a competitive
environment, there is always the fear of losing customers. While the primary
focus of each provider is to provide customer service satisfaction, preventing
customers from churning remains a huge challenge [1-2]. So that the services
and customers’ options also have become more comparable and more
competitive. This is the reason why subscriber loyalty tends to erode [3]. It is
very easy for customers to switch from a service provider to another without
losing anything. This causes the company to lose potential revenue/profit.
Therefore, companies have a hard time offering better services by introducing
new art applications and technologies to keep customers [4-6]. Before doing so,
it is necessary to identify customers who are likely to leave the company in the
near future, as losing them will result in a significant loss of profits for the
company.
Churn prediction analysis which is often used in all sectors, is one of the
applications of Machine Learning (ML). There should be detail information
about Business Support Systems (BSS) - Customer Reletionship Management
(CRM) and billing, Operations Support Systems (OSS), and Comlaint Work
Order (CWO) data to conduct these analyzes from the raw data [6].
In this paper, we review churn indicators strategy design for customer
churn prediction over existing research and as a more optimal option. We
present the detail about availability of derivative indicators that which is a better
indicators to clarify the churn prediction.
Data Collection
An accurate churn model is the core to predict potential customer churns.
The churn model is established by training and verifying the OSS, BSS and
customer CWO data collected from the live network.
In general, two approaches can be observed in the development of churn
prediction models. One is the theoretical approach and the other is the practical
approach. The theoretical approach is usually carried out by researchers who use
experimental observations, questionnaires over the Google drive plug-in, or
other methods as raw data. It is also possible to use simulative data. In the
course of theoretical research, it can be observed that there are gaps in the
indicators and certain data. The practical approach is to work on the data of real
telecom companies. Working on real data is usually a difficult and responsible
job. The difficulty is that operators have a lot of large amounts of data. It is a
complex process to identify the indicators that can affect the churn prediction
model and to separate these indicators and keep them in any basic system. The
purpose of our study was to determine the required indicators from real raw
data. At the same time, we tried to show the procedure for processing the
indicators until they are transferred to the churn prediction model.
In order to determine the indicators that will affect the churn prediction
model and to collect the data, it can be shown that the data source of telecom
companies in general is in three different areas. These are the OSS side, the BSS
side, and the customer CWO side. In general, it can be said that telecom
companies have 1500 indicators on the BSS side, more than 8000 indicators on
the OSS side and about 1000 indicators on the customer CWO side. It is known
that not all of these indicators are used or affected for the customer churn
prediction model. Therefore, it is necessary to review each indicator and identify
more appropriate indicators. You can see more detail in Fig. 1.
Fig. 1. Initial churn indicator framework.
OSS Data Collection
OSS data is collected differently by each operator. When we say data, the
primary source is data. The source of OSS data is signaling data, which
subscribers pass through various network elements, including both user data and
signaling data, and hold it together as part of a core element. The data access
and preprocessing subsystem receives and distributes data reported from probes,
converts the formats of the data, correlates x detail record (xDRs), such as xDRs
between the next generation network (NGN) and the circuit switched (CS)
domain, supplements subscriber information in xDRs, and handles raw signaling
queries. Obtains data collected and converts the data into formats that can be
recognized by the packet switched (PS) and the CS domains of the core network,
the long term evolution (LTE) network, the IP multimedia subsystem (IMS)
network, the wireless network and the NGN.
The OSS data is collected from core network element using some scripts
and codes for churn prediction model. Before configuration, the fields required
has been defined for collecting the OSS data. Thus, OSS data consists of a total
of 586 data, which contains 21 tables. The mentioned data can be grouped as
follows: MSISDN, TAC, SIM card, local carrier, location, NE information and
others. As for the services provided to the subscriber: CS basic service
indicators, PS & EPC basic service indicators and optional service indicators. As
an example, one can be shown below.
100*
BA A
CD
CD - Call Drop Rate After Answer(%)
A - GSM Call Drops After Answer(times) - Measure the number of Clear
Request messages received after the MSC serving the calling party sends the
Connect Ack message to the called party.
B - GSM Call Release After Answer(times) - Measure the total number of
Connect Ack messages sent by the calling party.
BSS Data Collection
The BSS data is provided by CRM and billing teams. The BSS data varies
depending on different operators. Typically, a comprehensive BSS data
collection generates a more accurate model. Therefore, the BSS data should be
collected as much as possible from the operator's live network. In this
configuration the data integrate basic customer information, charging
information, and service subscription information for customer churn prediction
model from the BSS.
Customer CWO Data Collection
The work order data records all the information exchanged between
customers and telecom operators, including customer complaints, service
consultation, and service handling. The work order data collected in this solution
contains:
Work order files: Contain services processed by Telecom Company.
Call details between customers and telecom company: Contain call
records between telecom company and customers.
Work order processing details: Contains the work order handling process
and result.
The OSS, BSS, and CWO Collection data is collected to database and
needed keep minimum data of 3 months for churn prediction.
Data Cleansing and Pre-processing
After collected data we need to perform the data cleansing and pre-
processing. Thus, it is necessary to ensure the consistency of these processes:
- To determine the necessary OSS&BSS derivative KPIs/KQIs.
- To develop the scripts and directs.
- To verify the data accuracy and integrity.
The OSS, BSS, and CWO, data that has been collected includes the noise
data and irrelevant data. The noise and irrelevant data has a negative effect on
the churn prediction modeling. The purpose of data cleansing is to remove the
noise data and irrelevant data from the source dataset. Examples of the data
cleansing are as follows:
De-duplicate the data. Some data, such as the customer record, is
duplicated in the collected data. Delete such data.
Delete the data with logic errors. A record that a customer's call drop
times are greater than the dialing times is generated due to the packet
losses. Delete such data.
Delete phone numbers which do not conform to the operator's number
specification. These phone numbers are error numbers that a customer has
dialed or some public numbers.
The above steps makes the data cleansing specifications and requirements.
After that, the necessary scripts and jobs are prepared using database
managment system (DBMS).
In general, the BSS&OSS&CWO data of 6 months are collected. The raw
data is collected by hour or day. To facilitate the data storage and system
operation, this data needs to be categorized as the files by day, week, and month.
Derivative Indicators
The churn tendency model is based on customer backgrounds, characters,
social contexts, and historical complaints. The historical complaint data is
obtained from complaint work orders, and customer background data is obtained
from the BSS customer profile table and consumption table. Indicators used for
the quality churn model include not only the standard KQIs provided by OSS
data, but also KQIs of different intervals (such as monthly, weekly, previous
day, and current day), hot spot cell KQIs, social network KQIs, and device type
KQIs. Therefore, the response indicator data needs to be derived from the
collected BSS and OSS data, and is used for modeling.
Indicators are determined to be derived based on the OSS and BSS low
layer data and customer experience, and provides this derived data to the churn
prediction model.
Now the last step is convert the operational definition to the network
indicators, match the indicators to the corresponding operational definition, find
out the churn indicators, analyze and enhance the indicators, verify whether they
can reflect the specific segment needs according to the user needs theory, if it
can reflect the operator service level and user experience, then keep, else
remove.
Finally construct the churn indicators set framework: User Persona
Indicator, Churn Motivation Indicator and Churn Warning Indicator
Conclusion
Churn indicators solution design give a concept to us how to do analyze
and churn prediction strategy using OSS, BSS, and CWO data in Telecom
Company. We have presented the optimal use of data science in customer churn
prediction and management by using a case study. We tried to keep any
indicators that would disturb the customer in the list of indicators. More
specifically, the paper presents techniques for churn indicators by useful way.
Customer churn indicators are explained in detail in theoretical, practical and
technical terms from raw data. Thus, even if the source of raw data is the same
in all telecom companies, the correct selection of the initial indicators will get to
a more accurate definition of the churn prediction.
References
1. Huang, B., Kechadi, M.T., Buckley, B.: Customer churn prediction in
telecommunications. Expert Syst. Appl. 39, 1414–1425 (2012)
2. Alae Chouiekh, El Hassane Ibn El Haj.(2020) Deep Convolutional
Neural Networks for Customer Churn Prediction Analysis.
International Journal of Cognitive Informatics and Natural
Intelligence, Volume 14, pp 1-16; doi:10.4018/ijcini.2020010101
3. Abbas Keramati, RuhollaJafari-Marandi Mohammed, Aliannejadi
ImanAhmadian and MahdiehMozaffari UldozAbbasi, "Improved
churn prediction in telecommunication industry using data mining
techniques", Applied Soft Computing, vol. 24, pp. 994-1012, 2014.
4. V. Umayaparvathi and K. Iyakutti, "A Survey on Customer Churn
Prediction in Telecom Industry: Datasets, Methods and Metrics,"
International Research Journal of Engineering and Technology
(IRJET), vol. 03, no. 04, April 2016.
5. Kamalraj, N., and A. Malathi. "A survey on churn prediction
techniques in communication sector." International Journal of
Computer Applications 64, no. 5 (2013).
6. Niyamaddin Taghiyev.(2020) “Churn prediction: indicators and model
design.” aztu.edu.az/azp/yubiley/az/main/main.jsp#
