ISSN: 2455-0345

International Journal of Volume 10, Issue 2, 2024

Telecommunication and Emerging Technologies July–December

DOI (Journal):10.37591/IJTET

https://journalspub.com/journal/ijtet

Review IJSCT

Machine Learning Approaches for Customer Churn

Prediction in Telecommunications
Anjali Banarase1, Ruchi Dave2,*, Vaishnavi Chavan3, Kajal Bhagwat4

Abstract
Customer churn prediction is a critical challenge for companies in the telecommunications industry.
The competition is intense, and acquiring new customers can be costly. Predicting when existing
customers might leave (churn) helps companies take proactive measures to retain them, reducing
churn rates and costs. By analyzing historical customer data, including demographics, service usage,
billing, and customer service interactions, this study aims to identify customers at risk of leaving their
current service provider. The goal is to develop a robust predictive model that can help telecom
companies take proactive steps in customer retention. To identify the most efficient model, the study
assesses the performance of several algorithms using criteria such as accuracy, precision, recall, and
others. Through thoughtful feature selection, preprocessing, and model evaluation, our research
showcases how machine learning can be a vital tool for identifying and reducing churn, leading to
cost savings and improved customer satisfaction. As the telecom sector becomes more competitive, the
ability to anticipate customer behavior and preempt churn will play a critical role in sustaining
profitability. By applying advanced predictive models, telecom providers can not only retain valuable
customers but also make data-driven decisions that drive business growth. The findings of this
research contribute to the growing body of knowledge on machine learning applications in the
telecom industry.

Keywords: Customer churn prediction, telecommunication industry, machine learning algorithms,

predictive modeling.

INTRODUCTION
In the very competitive telecom sector of today, customer retention is essential to long-term growth
and financial success. Telecom firms have a big issue from customer churn, the phenomenon where
customers move from one service provider to another.

Prediction of acquiring new customers far outweighs that of retaining existing ones, making churn
prediction a critical focus area. Leveraging the
power of machine learning (ML) algorithms,
*Author for Correspondence telecom companies can proactively identify
Ruchi Dave
E-mail: [email protected] potential churners and implement targeted
strategies to retain them. This research explores the
1–4
Student, Department of Electronics & Telecommunication use of ML methods for predicting customer
Engineering, Smt. Kashibai Navale College of Engineering,
Vadgaon, Pune, Maharashtra, India. attrition in the telecom industry. Through the lens
of data-driven analysis, this study contributes to
Received date: August 20, 2024
Accepted date: September 23, 2024
the arsenal of tools that telecom companies can
Published date: September 30, 2024 deploy to ensure customer retention and achieve
sustainable growth.
Citation: Anjali Banarase, Ruchi Dave, Vaishnavi Chavan,
Kajal Bhagwat. Machine Learning Approaches for Customer Customer churn prediction (CCP) is a critical
Churn Prediction in Telecommunications. International Journal
of Telecommunication and Emerging Technologies. 2024; task in the telecom industry, given the competitive
10(2):1–7. nature of the market and the substantial cost

© JournalsPub 2024. All Rights Reserved 1

Machine Learning Approaches for Customer Churn Prediction Banarase et al.

associated with acquiring new customers. This research explores the use of ML methods for
predicting customer attrition in the telecom industry. In this research, historical customer data,
including demographic information, usage patterns, billing details, and customer service interactions,
is utilized as the basis for predictive modeling.

There is great potential for the telecom industry to use ML algorithms for customer attrition
prediction. Because the telecommunications industry is so competitive, keeping consumers is essential
to sustaining profitability and market share. Businesses in this sector can gain a deeper understanding
of consumer behavior, spot possible churners, and lower churn rates by utilizing ML.

A key idea in ML is continuous learning, often known as lifetime learning, which stresses the
capacity to learn from new tasks while retaining information from earlier ones. Using past
experiences, this method enables models to adjust to a range of tasks in different areas.

Google DeepMind’s Elastic Weight Consolidation (EWC) is one efficient way to accomplish this.
Catastrophic forgetting, a prevalent issue where a model forgets previously learned information when
training on new tasks, is avoided with the use of EWC. It accomplishes this by determining how
significant various neural network weights are.

EWC modifies the model’s plasticity in a targeted manner by determining which weights are
essential to sustaining performance on previous tasks. This implies that the model can learn new tasks
without appreciably changing the key weights that correspond to previously learned information. Over
time, models become more adaptive and efficient because of this mechanism, which enables more
robust learning in dynamic contexts [8].

Because of the enormous costs involved, focusing on obtaining new consumers is becoming less
viable in the telco business. It is more economical to prioritize customer retention instead.
Consequently, efficient churn control has become crucial. This is recognizing clients who are in
danger, figuring out what they need, and putting plans in place to increase their loyalty. By doing this,
telecom businesses may lower employee churn, raise client satisfaction levels, and eventually boost
profitability [9].

By leveraging the insights gained from these algorithms, businesses can optimize their customer
retention strategies, reduce churn rates, and ultimately enhance customer satisfaction and profitability.
As the field of ML continues to advance, it opens new opportunities for telecom companies to stay
ahead in the competitive landscape.

Customer churn analysis is the process of identifying the operational measures that must be taken to
prevent customers who tend to cancel their subscriptions or services for a variety of reasons from
doing so. This process is particularly relevant in the financial, insurance, and telecommunications
sectors [10].

LITERATURE REVIEW
Using a hybrid model that integrates well-liked ML techniques, Khamlichi et al. [1] presented a
novel methodology for CCP in the telecom sector. They employed cross-validation to apply and
assess several models on a popular public dataset during the first stage of their research. They then
went on to create their hybrid model, which showed a notable gain in performance above separate
models. They found the best configurations by running large-scale simulations with different
combinations of parameters. Their findings show how much better the suggested model is than the
state-of-the-art ML methods, offering insightful information on churn prediction.

© JournalsPub 2024. All Rights Reserved 2

International Journal of Telecommunication and Emerging Technologies
Volume 10, Issue 2
ISSN: 2455-0345

To forecast customer turnover, Kaur et al. [2] investigated several machine-learning models on a
bank dataset. These models included Logistic Regression (LR), Decision Tree (DT), K-Nearest
Neighbor (KNN), and Random Forest (RF). Accuracy, recall, and other performance criteria are used
in the paper’s comparison of these models. This analysis helps to discover the most successful
algorithm for predicting customer churn, highlighting the strengths and limitations of each model in
different scenarios, ultimately directing the choice of predictive techniques in the financial sector.

In the telecom sector, Li et al. [3] presented a combination model for CCP that makes use of a
gradient-boosting decision tree (GBDT) and a one-dimensional convolutional neural network
(1DCNN). Customer data are first fed into the 1DCNN by the model, which uses one-dimensional
convolution to automatically extract features. The fully linked layer is then used to carry out the churn
forecast. When compared to traditional approaches, this integrated strategy is effective in predicting
customer churn, as evidenced by the considerable improvement in recall rate and F1 score observed in
experimental results on two publicly available telecom datasets.

In the telecommunications sector, where customers regularly switch to other service providers, Tsai
and Prasad [4] emphasize the vital significance of client retention for corporate viability. As they
point out in their paper, “An Intelligent Customer Churn Prediction and Response,” anticipating
customer churn is essential to lowering attrition and raising client lifetime value. Businesses may
lower attrition rates by implementing focused interventions after correctly identifying the customers
most likely to leave. They provide insightful information for proactive client retention tactics with
their study on the application of intelligent systems to forecast and address customer turnover.

Robotyshyn and Malyar [5] presented a study at the 2020 IEEE 2nd International Conference on
System Analysis & Intelligent Computing (SAIC) that discusses CCP as a traditional and increasingly
relevant application of ML techniques. The authors highlight the growing importance of ML models
for firms looking to retain consumers by focusing on their usage in forecasting churn. Their research
sheds light on the potential of several ML techniques to raise the precision and efficacy of churn
predictions and offers insightful information on how these techniques might be used in real-world
business contexts.

In their paper titled “Using XGBoost (DNN and MLP), a Model for Predicting Customer Churn,”
presented at the 2020 International Conference on Computer Engineering and Application (ICCEA),
Tang and Xia [6] investigate cutting-edge ML techniques for CCP. The authors draw attention to the
difficulties that conventional deep neural networks (DNNs) have when processing numerical
information, which is essential for producing precise churn forecasts. They suggest a hybrid model
that combines XGBoost and multilayer perceptrons (MLP) to address this. By combining the
advantages of neural networks and tree-based structures, the model handles numerical data efficiently
and provides a reliable solution for churn prediction in a variety of areas. This leads to a significant
improvement in predictive performance.

The cost of acquiring new customers far outweighs that of retaining existing ones, making churn
prediction a critical focus area. Through the lens of data-driven analysis, this study contributes to the
arsenal of tools that telecom companies can deploy to ensure customer retention and achieve
sustainable growth.

In their work, Venkatesh and Jeyakarthic et al. [7] discuss the growing significance of CCP and
highlight the growing use of ML models to address this issue. Advances in cloud computing and the
Internet of Things (IoT) have made it possible for organizations to gather large amounts of customer
data to support CCP initiatives. The authors of the paper suggest using a support vector machine
(SVM) model in conjunction with an optimum genetic algorithm (OGA) to predict churn. By
optimizing the SVM’s performance, a double-chain quantum genetic algorithm is used in the

© JournalsPub 2024. All Rights Reserved 3

Machine Learning Approaches for Customer Churn Prediction Banarase et al.

development of the OGA, providing a more precise and effective method for CCP.

“Jafar & Aljoumaa [11], 2019: Customer Churn Analysis and Prediction in Telecom Industry using
Machine Learning”. One of the most popular study areas in the telecom business in recent years is
CCP. It entails identifying clients who are most likely to discontinue a service subscription.

BLOCK DIAGRAM
The block diagram for CCP includes several key processes, illustrated in Figure 1. Each block in
the diagram represents a distinct operation, which is described below.

Figure 1. Block diagram of churn prediction.

• User: User role could involve data preprocessing, feature engineering, model selection, and
evaluation.
• Input as text: The input as a text box could be used to gather textual information from customers.
This could include feedback, comments, or any other text-based data that might be relevant for
predicting churn. From Figure 1 we can see, the text box allows customers to express their
thoughts and concerns, which can be valuable in understanding their sentiments and identifying
potential churn indicators. This text data can then be processed and incorporated as features in the
ML model to improve churn prediction accuracy.
• Preprocessing: The act of cleaning and transforming the data before supplying it to the ML model
is referred to as the “preprocessing box”. The system architecture is illustrated in Figure 1, which
outlines the flow of data through the various steps of preprocessing. This includes operations like
scaling numerical characteristics, encoding categorical variables, and managing missing data. By
ensuring that the input is in an appropriate format, preprocessing helps the model learn from the
data and produce correct predictions.
• Train dataset: To teach ML models to process data and make correct predictions, training data are
a sizable dataset. Typically, this means imparting knowledge to prediction models using learning
algorithms on how to extract features pertinent to certain business objectives.
• Feature extraction: Using the existing data, we generate pertinent characteristics in the feature
extraction box that aid in churn prediction. These features capture customer behavior, patterns,
and characteristics that may be indicative of potential churn. For example, we can extract features
such as customer tenure, usage frequency, average transaction amount, and customer
demographics. These characteristics give the ML model useful information that helps it anticipate
client turnover accurately.
• Classification SVM/RF/DT: The classification SVM/RF/DT box refers to the step where we select
and train the appropriate classification algorithms for predicting churn. SVM, RF, and DT are
popular algorithms used for classification tasks.
• Prediction of churn: Using the learned ML model, we can forecast the likelihood of a customer
churning or not in the churn box prediction. The model processes the pertinent features and

© JournalsPub 2024. All Rights Reserved 4

International Journal of Telecommunication and Emerging Technologies
Volume 10, Issue 2
ISSN: 2455-0345

predicts or estimates the likelihood that a customer will leave. By putting focused retention tactics
into place, this forecast can assist organizations in being proactive in retaining customers and
reducing attrition.
• Display output to the user: The display output to the user box refers to the step where we present
the predicted results to the user. This could be in the form of a report, dashboard, or any other
visual representation that communicates the churn predictions effectively. By presenting the
predictions clearly and understandably, businesses can make informed decisions and take
appropriate actions to mitigate churn and retain valuable customers.

DATA FLOW DIAGRAM

The Data Flow Diagram (DFD0) illustrates the data flow in our system by using a circle to
represent our system and a rectangle to represent input and output. The processes involved in churn
prediction, such as feature extraction and model training, are shown in Figure 2. In DFD1, we display
the actual input and output of the system. Text or images are the input, and rumors are identified as
the output. Similarly, in DFD2, we display the user and administrator operations.

Figure 2. Data processing.

Hardware Specification
• System: Intel I5 Processor.
• Hard Disk: 40 GB.
• RAM : 8 GB.

© JournalsPub 2024. All Rights Reserved 5

Machine Learning Approaches for Customer Churn Prediction Banarase et al.

Software Specification
• Operating system: Windows 10.
• Coding language: Python.
• IDE: Spyder.
• Database: SQLite.

ADVANTAGES
● Businesses can use targeted retention efforts by using churn prediction to identify customers
who are likely to leave.
● It is frequently less expensive to keep current clients than to find new ones.
● Churn prediction models enable businesses to personalize marketing campaigns, offers, and
incentives for specific customer segments, improving the effectiveness of marketing efforts.
● This leads to higher rates of client retention.

DISADVANTAGES
● Churn prediction models enable businesses to personalize marketing campaigns, offers, and
incentives for specific customer segments, improving the effectiveness of marketing efforts.
● Gathering and analyzing customer data raises privacy concerns. Companies need to make sure
that laws like the CCPA and GDPR, which protect personal data, are followed.
● Churn prediction model construction and upkeep can be difficult and resource intensive. It
might call for knowledge of ML and data science.

APPLICATIONS
● Churn prediction is a tool that telecom companies can use to determine which customers are
most likely to migrate to a competitor.
● Streaming platforms, like Netflix and Spotify, use churn prediction to understand customer
behavior and preferences.
● Online retailers can predict which customers are at risk of abandoning their shopping carts or
not making repeat purchases.

CONCLUSION AND FUTURE SCOPE

● The future scope of CCP using ML algorithms is promising and involves several potential
advancements and areas of development.
● Advanced ML models: Exploration of more sophisticated ML models, including deep learning
architectures, to capture intricate patterns and dependencies in customer behavior data.
● Explainable AI (XAI): Enhanced emphasis on creating interpretable and accurate models,
meeting the requirement for openness in decision-making procedures.
● Personalization and contextual analysis: Enhancement of models to provide more personalized
predictions by considering individual customer preferences, behavior patterns, and contextual
factors.
● Dynamic and adaptive modeling: Development of models that can dynamically adapt to
evolving customer behaviors over time, allowing for more accurate predictions in dynamic
business environments. 32 Department of E&TC Engineering, SKNCOE, Pune-413326.
● Integration with big data and streaming analytics: Integration with big data technologies and
streaming analytics for handling large volumes of data in real-time, especially in industries with
high-frequency customer interactions.
● Cross-channel and multimodal analysis: Exploration of cross-channel analysis and
incorporation of multimodal data (e.g., merging organized and unstructured data) to get a deeper
knowledge of interactions with customers).
● Hybrid models and ensemble techniques: Experimentation with hybrid models that combine the

© JournalsPub 2024. All Rights Reserved 6

International Journal of Telecommunication and Emerging Technologies
Volume 10, Issue 2
ISSN: 2455-0345

strengths of different algorithms and the use of ensemble techniques to further boost predictive
accuracy.
● Blockchain for data security: Using blockchain technology to improve client data security and
integrity addressing concerns related to privacy and data protection.

ACKNOWLEDGMENT
The combination of sophisticated algorithms, thoughtful feature engineering, and rigorous model
evaluation can empower telecom companies to make informed decisions and foster customer loyalty
in an increasingly competitive landscape.

REFERENCES
1. Khamlichi FI, Zaim D, Khalifa K. A new model based on global hybridization of machine
learning techniques for “customer churn prediction. In: 2019 Third International Conference on
Intelligent Computing in Data Sciences (ICDS). IEEE. 2019 Oct 28. pp. 1–4.
2. Kaur I, Kaur J. Customer churn analysis and prediction in banking industry using machine
learning. In: 2020 Sixth International Conference on Parallel, Distributed and Grid Computing
(PDGC). IEEE. 2020 Nov 6. pp. 434–437.
3. Li S, Xia G, Zhang X. Customer churn combination prediction model based on convolutional
neural network and gradient boosting decision tree. In: Proceedings of the 2022 5th International
Conference on Algorithms, Computing and Artificial Intelligence. 2022 Dec 23. pp. 1–6.
4. Tsai TY, Lin CT, Prasad M. An intelligent customer churn prediction and response framework.
In: 2019 IEEE 14th International Conference on Intelligent Systems and Knowledge Engineering
(ISKE). IEEE. 2019 Nov 14. pp. 928–935.
5. Malyar M, Robotyshyn MM, Sharkadi M. Churn prediction estimation based on machine learning
methods. In: 2020 IEEE 2nd International Conference on System Analysis & Intelligent
Computing (SAIC). IEEE. 2020 Oct 5. pp. 1–4.
6. Tang Q, Xia G, Zhang X, Long F. A customer churn prediction model based on XGBoost and
MLP. In: 2020 International Conference on Computer Engineering and Application (ICCEA).
IEEE. 2020 Mar 18. pp. 608–612.
7. Venkatesh S, Jeyakarthic M. An optimal genetic algorithm with support vector machine for
cloud-based customer churn prediction. In: 2020 International Conference on System,
Computation, Automation and Networking (ICSCAN). IEEE. 2020 Jul 3. pp. 1–6.
8. Sarnen F, Suyanto S, Rismala R. Lifelong learning for dynamic churn prediction. In: 2020
International Conference on Data Science and Its Applications (ICoDSA). IEEE. 2020 Aug 5. pp.
1–5.
9. Wu S, Yau WC, Ong TS, Chong SC. Integrated churn prediction and customer segmentation
framework for telco business. IEEE Access. 2021;9:62118–36.
10. De S, Prabu P. A representation-based query strategy to derive qualitative features for improved
churn prediction. IEEE Access. 2023;11:1213–23.
11. Al-Zuabi IM, Jafar A, Aljoumaa K. Predicting customer’s gender and age depending on mobile
phone data. J Big Data. 2019;6(1):18.

© JournalsPub 2024. All Rights Reserved 7