The Investigation of the Advancement for Machine Learning-Based

Telecommunication Customer Churn Prediction

Yi Zhang

College of Letters & Science, University of Wisconsin Madison, Madison, U.S.A.

Keywords: Machine Learning, Customer Churn, Deep Learning.

Abstract: It is important for companies to predict customer churn, as it helps reduce losses, and machine learning is

crucial to this process, making its study extremely important. Ensemble Methods, Logistic Models,

Classification Trees, and Nearest Neighbor Algorithms are all traditional machine learning methods, but they

have inherent drawbacks. This encompasses overfitting, a situation where models achieve high accuracy on

training data but struggle to perform effectively on unseen data; model complexity, making complex models

difficult to interpret and maintain; and inefficiencies in handling large-scale data, often leading to poor

performance with vast amounts of data. Conversely, advancements in deep learning highlight its strengths

and areas where it surpasses traditional methods. This study offers an extensive perspective on telco churn

prediction and artificial intelligence, delving into possible concerns like data protection, which aims to protect

user data from exploitation. To address these challenges, several solutions are proposed, including

collaborating with experts to ensure the accuracy and reliability of AI models, implementing stronger security

measures to protect sensitive information, and utilizing techniques to mitigate data scarcity issues. Overall,

this work offers an excellent review of the telco churn prediction field, highlighting key advancements and

proposing solutions to existing challenges.

1 INTRODUCTION

Telco churn is when a telecommunications company

loses customers. Customers may decide to cancel

their current service and switch to another company

if they are dissatisfied with the current service or if

another company offers better deals and packages. In

the telecom industry, competition is fierce. Retaining

existing customers is more cost-effective than

acquiring new ones. Customer churn directly impacts

a company's revenue and market position, so telecom

companies pay close attention to this issue. Therefore,

accurately identifying customer churn to help

companies reduce losses is crucial.

In the last few years, Artificial Intelligence (AI)

has gradually developed in the society and found

widespread applications in the economy, including

significant increases in robotics, AI startups, and

patent numbers (Furman & Seamans, 2019). And at

the same time, AI technology has exceeded human

capabilities in several domains, including image

recognition and speech recognition. AI is widely used

https://orcid.org/0009-0008-0144-0573

in search engines, health monitoring, stock analysis

etc. Investment in AI is rapidly increasing worldwide,

with governments actively promoting AI research.

The future of AI may involve cyclical periods of

growth and decline or rapid development to a level

that surpasses human intelligence (De Spiegeleire et

al., 2017). AI has been instrumental in the

advancement of predictive analytics methods,

resulting in the development of numerous predictive

models. Supervised learning uses labeled data for

training, such as classification trees and margin-based

classifiers, and is frequently employed for

categorization and regression tasks. Unsupervised

learning handles unlabeled data, like Dimensionality

reduction technique and k-means clustering, and is

mainly used to discover patterns and structures in the

data. These algorithms are applicable to different

types of data and tasks, with widespread applications

ranging from online shopping recommendations to

photo updates on social networks (Mahesh, 2020).

Machine learning analyzes data to help businesses

and governments make better decision. It plays a

Zhang, Y.

The Investigation of the Advancement for Machine Learning-Based Telecommunication Customer Churn Prediction.

DOI: 10.5220/0013206400004568

In Proceedings of the 1st International Conference on E-commerce and Artiﬁcial Intelligence (ECAI 2024), pages 56-60

ISBN: 978-989-758-726-9

crucial role in cybersecurity by identifying patterns to

better detect malware and cyber-attacks.

Additionally, machine learning helps develop smart

applications that understand user behavior and

provide personalized recommendations and services

(Sarker, 2021). One of the tasks that has received

significant attention is predicting customer churn in

the telecommunications industry. There has been

extensive research on this topic in the past. A churn

prediction model based on the ensemble methods

algorithm has been developed by researchers from

various universities and research institutions in

Pakistan and South Korea. This model analyzes

subscriber attrition in the telecommunications

industry, identifying the causes and behavioral

patterns of churn. The aim is to help telecom

companies reduce customer churn, optimize their

Customer Relationship Management (CRM) systems,

and enhance their profitability (Ullah et al., 2019).

And at the same time, the research team from

Mangalam College of Engineering in India developed

a churn prediction model utilizing deep learning and

ensemble methods algorithms. This model is able to

examine client retention issues in the

telecommunications sector, identifying the reasons

for churn and detecting associated behavioral patterns

(Andrews et al., 2019). In recent years, many new

algorithms have emerged, making it necessary to

conduct a comprehensive review of this area. The rest

of this paper is structured into three main sections:

methods, discussion, and conclusion.

This paper will firstly detail the various methods

used by others to predict telecom customer churn, and

then discuss the strengths and weaknesses of these

methods and provide insights into future directions.

Finally, this paper will summarize the key points of

the entire paper.

2 METHOD

2.1 The Introduction of the Machine

Learning Workflow

Machine learning algorithms typically involve the

following steps: data collection, preprocessing,

model building, training, and testing. First, collect a

large amount of high-quality data. Then, clean and

standardize the data, extract useful features, and

reduce dimensionality. Next, choose the appropriate

algorithm based on the specific problem, such as

classification trees, and Margin-based classifier for

supervised learning; k-means algorithms for

unsupervised learning; and transductive support

vector machines for semi-supervised learning. Train

the model with the training data, fine-tuning

parameters and optimizing to enhance performance.

Finally, test the model's accuracy with test data by

Assessing performance indicators like correctness,

specificity, sensitivity, and F1 measure on the new

data. Different algorithms have their own strengths

and weaknesses: supervised learning is suitable for

small, well-labeled datasets, while unsupervised

learning and deep learning perform better on large-

scale datasets. Machine learning has numerous

applications in contemporary life, from shopping

recommendations to photo recognition on social

networks (Mahesh, 2020).

2.2 Traditional Artificial Intelligence

Techniques-based Prediction

2.2.1 Ensemble Methods

Ensemble methods is a machine learning algorithm

employed for making predictions. It generates

multiple classification trees and combines their

results to enhance prediction accuracy (Probst et al.,

2019). The ensemble methods algorithm has

numerous applications in the telecommunications

sector. It analyzes customer data by generating many

classification trees and combines the results of these

trees to make predictions. This method enables more

precise identification of customers likely to churn.

The findings show that the ensemble methods

algorithm can accurately predict customer churn with

an accuracy of 88.63%, which is extremely

advantageous for telecom companies in developing

strategies to retain their customers (Ullah et al.,

2019). And this method is highly effective in handling

large datasets, offering excellent classification

accuracy and fast processing. Experimental results

show that it achieves a prediction accuracy of 91.4%,

making it very useful for telecom companies in

developing customer retention strategies

(Abdulsalam et al., 2022).

2.2.2 Logistic Model

Logistic model is a statistical technique that predicts

the likelihood of a binary outcome based on one or

more predictor variables. This method is frequently

applied in fields such as social sciences and medical

research to analyze relationships between dependent

and independent variables, allowing researchers to

make informed predictions and decisions (Fletcher &

Islam, 2019). In telecom customer churn prediction,

logistic model is extensively utilized because of its

The Investigation of the Advancement for Machine Learning-Based Telecommunication Customer Churn Prediction

simplicity and ease of interpretation. It provides

easily understandable results and probability outputs,

is suitable for mixed data types, and has high

computational efficiency.

2.2.3 Classification Trees

Classification trees are non-parametric supervised

learning algorithms employed for both classification

and regression problems. They make decisions by

recursively dividing the dataset into subsets. They

offer high interpretability, can handle both continuous

and discrete data, and discover non-linear

relationships between attributes (Fletcher & Islam,

2019). Classification trees have extensive

applications in predicting customer churn for telecom

companies. By analyzing customer data and

recursively partitioning it into subsets, classification

trees effectively predict which customers are likely to

churn. Experimental results show that classification

trees excel in classification accuracy, achieving a 93%

accuracy rate, which is very helpful for telecom

companies in devising customer retention strategies

(Hassonah et al., 2019).

2.2.4 Nearest Neighbor Algorithms

The nearest neighbor algorithms (KNN) algorithm is

a non-parametric supervised learning approach that

categorizes new data points by calculating the

distances between them and all points in the training

dataset. In telecom churn prediction, the KNN

algorithm involves collecting and standardizing

customer data, training the model using historical

data, followed by calculating the distances between

new customers and all points in the training dataset.

By identifying the KNN, the algorithm predicts

whether a customer will churn based on the

classifications of these neighbors, helping telecom

companies identify and retain potential churners

(Hassonah et al., 2019).

2.3 Deep Learning-Based Prediction

2.3.1 Artificial Neurons

The Artificial Neural Network (ANN) is a

computational method that mimics the way the

human brain works. It improves prediction accuracy

by learning from data. ANN include several layers:

input levels, hidden levels, and output levels. The

intermediate layers process complex information,

thereby improving the accuracy of the predictions.

ANN also have significant applications in predicting

telecom customer churn. By using a dataset

containing 3333 samples and 21 attributes and

applying a feature selection algorithm called Relief-

F, the authors identified the most useful features for

prediction. These features were then used to train and

test the ANN model, which achieved a prediction

accuracy of 93.88%, outperforming other methods.

This demonstrates that ANN is highly effective in

handling complex data and accurately predicting

whether customers will churn (Abdulsalam et al.,

2022).

3 DISCUSSIONS

Lately, machine learning has gradually evolved from

traditional methods to deep learning. Traditional

machine learning methods have revealed many

shortcomings, particularly when handling large and

complex datasets, making these shortcomings

especially prominent in today's data-driven telco

churn prediction applications. Deep learning, with its

powerful data processing capabilities and complex

pattern recognition, enables more accurate churn

predictions, helping telecom companies to more

effectively devise customer retention strategies.

Firstly, when the data volume is exceptionally

large, traditional machine learning methods exhibit

poor computational efficiency and often struggle to

handle large-scale datasets. For example, when

dealing with millions or even billions of data records,

traditional algorithms often have high computational

complexity, leading to long processing times and an

inability to meet practical application demands.

Secondly, traditional machine learning methods

have high requirements for data quality. When data is

missing, noisy (e.g., contains erroneous information),

or incomplete, the performance of traditional methods

can significantly degrade, making it difficult to

handle these issues effectively. Furthermore,

traditional models have limited generalization

capabilities across different datasets. Many

traditional models perform well on training data but

show a significant drop in performance on new data.

Lastly, traditional machine learning methods have

limited capability in handling complex problems. For

instance, in the logistic model, it assumes a linear

relationship and may not capture complex customer

behaviors, making its predictive performance

potentially inferior to more sophisticated algorithms.

However, Deep learning models typically have

more significant advantages compared to traditional

methods. Such nonlinearities can be effectively

captured by them, resulting in greater accuracy,

particularly when managing large datasets and

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence

complex tasks like visual recognition, audio

recognition, and language interpretation. Deep

learning can handle vast amounts of data

exceptionally well, and the more data it processes, the

better it performs, making it particularly

advantageous in big data environments. Additionally,

deep learning can be applied across multiple fields,

not just limited to a single domain.

In the telecommunications sector, AI is essential

for forecasting customer churn. By analyzing vast

amounts of customer data, AI can pinpoint those

likely to churn, assess the risk, and enable telecom

companies to take proactive steps to reduce churn

rates, thus increasing potential profits. But there are

also some limitations. AI needs a lot of data to learn

and work properly. If there isn't enough data or the

data quality is poor, AI's performance can suffer.

Moreover, collecting and processing this data is both

time-consuming and expensive.

Additionally, in most cases, AI can only provide

a result without explaining why it came to that result.

However, in practical applications, such as in

healthcare and legal fields, decision-makers need to

understand the reasons and processes behind these

predictions.

Training complex AI models requires powerful

computers and a lot of time. This can be a significant

burden for many small companies and research

institutions. AI performs well on familiar data, but it

may struggle with new, unseen data. This means that

AI can sometimes be unreliable in practical use.

Furthermore, highly autonomous AI systems can

pose safety risks. If an AI system behaves

unexpectedly or is maliciously exploited, it can lead

to serious consequences. Ensuring that AI systems

operate safely in all situations is a significant

challenge.

Recently, with the continuous advancement of

technology, some possible methods can be considered

to address the existing issues mentioned above. To

address the issue of AI's lack of interpretability and

reliable, companies can collaborate with domain

experts. These experts can provide business

knowledge and feedback, helping to select important

features, validate, and interpret AI model predictions,

ensuring that the models align better with actual

business needs and making the results more

trustworthy and easier to understand. Furthermore, to

address the issues of data scarcity and long training

times, which can lead to high costs or render some

domains impractical, transfer learning offers an

effective solution. Transfer learning plays a crucial

role in transferring knowledge between different yet

related domains. It reduces the reliance on a large

amount of labeled data in the target domain, thus

enhancing the performance of models within that

domain. It also demonstrates broad application

prospects in addressing data scarcity in real-world

scenarios and optimizing machine learning model

performance (Zhuang et al., 2020). Meanwhile, to

address the issue of AI's lack of privacy, Federated

Learning (FL) can effectively solve this problem. FL

holds significant importance and has vast applications

across various fields, including industrial engineering

and healthcare. FL is a decentralized machine

learning technique designed to enable multiple

devices or organizations to train models together

without disclosing raw data, thereby protecting data

protection and security. Its primary goal is to address

data silos and privacy issues by allowing multiple

clients to participate in model training without

centralized data processing. The key characteristics of

FL include decentralization and data privacy

protection, making it particularly important in

scenarios where data privacy and decentralized data

processing are crucial (Li et al., 2020).

4 CONCLUSIONS

In this study, both traditional machine learning and

deep learning are analyzed in depth. Several major

methods of traditional machine learning and their

inherent drawbacks, such as overfitting, model

complexity, and limitations in handling large-scale

data, are explored. Additionally, advancements in

deep learning are delved into, highlighting its

strengths and areas where it surpasses traditional

methods. The potential applications of AI across

various domains are examined, providing a

comprehensive outlook on the future of the field.

Current potential issues, including data privacy,

security, algorithmic bias, and ethical concerns, are

analyzed. To address these challenges, a range of

solutions is proposed. These solutions include

improved data processing techniques to ensure more

accurate and reliable AI models, the implementation

of more robust security measures to protect sensitive

information, and the development of fairer algorithm

designs to mitigate bias and promote ethical AI

practices. Through these efforts, valuable insights and

guidance for the continued development and

responsible deployment of artificial intelligence

technologies are aimed to be provided.

The Investigation of the Advancement for Machine Learning-Based Telecommunication Customer Churn Prediction

REFERENCES

Abdulsalam, S. O., Ajao, J. F., Balogun, B. F., & Arowolo,

M. O. 2022. A churn prediction system for

telecommunication company using random forest and

convolution neural network algorithms. EAI Endorsed

Transactions on Mobile Communications and

Applications, 7(21).

Abdulsalam, S. O., Arowolo, M. O., Saheed, Y. K., &

Afolayan, J. O. 2022. Customer churn prediction in

telecommunication industry using classification and

regression trees and artificial neural network

algorithms. Indonesian Journal of Electrical

Engineering and Informatics (IJEEI), 10(2), 431-440.

Andrews, R., Zacharias, R., Antony, S., & James, M. M.

2019. Churn prediction in telecom sector using machine

learning. International Journal of Information, 8(2).

De Spiegeleire, S., Maas, M., & Sweijs, T. 2017. Ai –

Today And Tomorrow. In Artificial Intelligence and

The Future of Defense: Strategic Implications for

Small- And Medium-Sized Force Providers (pp. 43–59).

Hague Centre for Strategic Studies.

http://www.jstor.org/stable/resrep12564.8

Fletcher, S., & Islam, M. Z. 2019. Decision tree

classification with differential privacy: A survey. ACM

Computing Surveys (CSUR), 52(4), 1-33.

Furman, J., & Seamans, R. 2019. AI and the

Economy. Innovation policy and the economy, 19(1),

161-191.

Hassonah, M. A., Rodan, A., Al-Tamimi, A. K., &

Alsakran, J. 2019, November. Churn prediction: A

comparative study using knn and decision trees.

In 2019 Sixth HCT Information Technology Trends

(ITT) (pp. 182-186). IEEE.

Li, L., Fan, Y., Tse, M., & Lin, K. Y. 2020. A review of

applications in federated learning. Computers &

Industrial Engineering, 149, 106854.

Mahesh, B. 2020. Machine learning algorithms-a

review. International Journal of Science and Research

(IJSR). [Internet], 9(1), 381-386.

Probst, P., Wright, M. N., & Boulesteix, A. L. 2019.

Hyperparameters and tuning strategies for random

forest. Wiley Interdisciplinary Reviews: data mining

and knowledge discovery, 9(3), e1301.

Sarker, I. H. 2021. Machine learning: Algorithms, real-

world applications and research directions. SN

computer science, 2(3), 160.

Ullah, I., Raza, B., Malik, A. K., Imran, M., Islam, S. U., &

Kim, S. W. 2019. A churn prediction model using

random forest: analysis of machine learning techniques

for churn prediction and factor identification in telecom

sector. IEEE access, 7, 60134-60149.

Zhuang, F., Qi, Z., Duan, K., Xi, D., Zhu, Y., Zhu, H., ... &

He, Q. 2020. A comprehensive survey on transfer

learning. Proceedings of the IEEE, 109(1), 43-76.

ECAI 2024 - International Conference on E-commerce and Artiﬁcial Intelligence