Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

Customer Churn Prediction

Lavina Anand Parulekar and Sanika Abasaheb Sardesai Prajkta Shriram Jamsandekar and Sampada Sanjay Parkar
SSPM’s College of Engineering SSPM’s College of Engineering
Computer Engineering Computer Engineering
Mumbai University Mumbai University

Sawant S.P (Prof.)

SSPM’s College of Engineering
Computer Engineering
Mumbai University

Abstract:- In today’s highly competitive business I. INTRODUCTION

landscape, customer retention revenue preservation,
customer experience improvement, and marketing Customer churn refers to the problem where customers
optimisation are critical factors for sustained growth and stop doing business with a company or organization. It is
profitability. Customer churn prediction is discontinuing a critical thing for businesses to understand and predict. It
their services or purchases, which presents a significant is the rate at which customers stop doing business with a
challenge for businesses across various industries. This particular company. It’s a critical thing because retaining
project focuses on developing a predictive model to expect existing customers is sometimes more costly than acquiring
customer churn in the banking sector using machine new customers. This project aims to tackle the high customer
learning techniques. The project involves the collection churn rate in the Banking Sector. In the banking industry,
and analysis of historical customer data, confined account retaining existing customers is sometimes more costly than
activity, transaction history, demographic information, acquiring new ones. Loyal, long-term customers use multiple
and customer service interactions. By extracting the right services offered by the bank, such as savings accounts, loans,
features from this data, a machine learning model is credit cards, and investment products. Losing customers can
trained to forecast which bank customers are at the result in a significant loss of revenue. Due to the high
highest risk of churning. A critical step in this study customer churn rate, industries have to face problems like
was the selection of relevant features that influence loss of revenue, reduced profitability, and damage to the
customer churn. Feature selection was guided by domain brand’s reputation. We are doing this project to develop
knowledge and feature importance analysis. The different a predictive model that can identify customers at risk of
classifiers were used and then trained on the training churning. It is a classification model. The limitations of this
dataset further ensuring the model’s optimal performance. predictive model are data availability, time, and resources.
The model’s performance is assessed through various The accuracy of the churn prediction model depends on the
evaluation metrics, including accuracy, precision, and Quality of data, representativeness of data, fewer variations
recall. Additionally, the project explores a model in data, feature selection, dimensionality reduction, time, etc.
illustration to uncover the influential factors contributing We used Random forest, KKN, XG-Boost, Naive Bayes, etc.
to customer churn within the banking context. This to check the accuracy of the model on the dataset. From
project’s outcomes can empower banks to take proactive that, we observed that random forest gives higher accuracy. To
measures in retaining customers, enhancing their overall implement this project we were required to implement certain
experience, and thereby preserving revenue streams. By steps like Data collection, selection of specific features to train
addressing customer churn, banks can foster long-term the model, model building, training of the model, validation
relationships, reduce customer acquisition costs, and boost of the model’s result, prediction of newly incoming data.
their competitiveness in the financial industry. The results The importance of the customer churn prediction is that it
of this project are expected to assist businesses in helps to Predicting customer churn that allows companies to
proactively retaining customers by targeting those at the proactively take actions to retain their valuable customers,
highest risk of churning. Ultimately, reducing customer minimize revenue loss, and improve customer satisfaction and
churn can lead to increased customer satisfaction, revenue, loyalty.
and long-term business sustainability.

IJISRT24MAR1246 www.ijisrt.com 2878

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

II. RELATED WORK system based on data analytics and machine learning
techniques. This system should accurately identify customers
Several studies shows that different techniques are used at risk of churning and provide actionable measures to
tofind out Customer Churn Rate in different fields: implement retention strategies. Developing a predictive model
 Amol, Chole,in 2023 evaluated Random Forest (RF), capable of accurately identifying customers who are at risk of
Support Vector Machine (SVM), k-Nearest Neighbors churn in the future. This involves analyzing historical
(KNN), and Convolutional Neural Network (CNN) customer data, transaction records, interaction history, and
algorithmsfor predicting NonChurn customers. other relevant information to recognize churn patterns.
 Ziyu Zhu,in 2023 study explored Logistic Regression and
Random Forest algorithms, reporting AUC values of IV. KEY HIGHLIGHTS
78%and 69%˙
 Vani Haridasan ,in 2023 study introduced the Arithmetic  Retention
Optimization Algorithm (AOA) and Stacked Bidirectional The primary goal is to retain valuable customers. By
Long Short-Term Memory (SBLSTM), demonstrating the identifying those at risk of churning, a business can implement
effectiveness of the AOASBLSTM model in classifying strategies to get them back.
CR and NCR classes
 Micheal Olaolu Arowolo , in 2022 study compared  Revenue Preservation
CNN and Random Forest classifiers, reporting prediction Churn prediction helps businesses protect their revenue
accuracies of 94% and 91 %, for forecasting the churn rate by preventing customer loss. Retaining existing customers is
in telecom companies. often more cost-effective than acquiringnew ones.
 Dr. K. Geetha ,in 2022 study investigated Decision Trees,
Random Forests, and XGBoost algorithms, capable of  Customer Experience Improvement
efficiently excluding non-essential data and forecasting the Understanding why customers churn provides the areas
training model ahead of time. where improvements are needed. This information can guide
 Sena Kasim ,in 2022 research identified Logistic product and serviceenhancements.
Regression, Support Vector Machine, Decision Tree,
Random Forest, XGBoost, and K-NN algorithms,  Marketing Optimization
indicating the suitability of ten features for churn analysis. Churn prediction enables businesses to allocate
 Edemealem Desalegn Kingawa,in 2022 study explored marketing resources more efficiently. So that companies can
Logistic and Multiple Regression, SVM, DNN, and more focus on retaining high-risk customers and serving the
Random Forest algorithms, ultimately selecting a Deep loyal ones.
Neural Network model for prediction, with a
recommendation to train and test it on a large dataset V. PROPOSED SYSTEM
 Sulim Kima,in 2022 study focused on Decision Trees for
churn prediction in e-commerce, comparing original data A proposed system for customer churn prediction
with predicted results to accurately identify churners, involves a detailed approach to identify and predict customers
highlighting how advancements in e-commerce have who are likely to leave a bank or stop using its services.
provided customers with numerous purchasing options.
 Dr. S.B. Kishor,in 2022 study utilized Deep Learning
(DL), Artificial Neural Networks (ANN), and
Convolutional Neural Networks (CNN), preprocessing raw
data to convert categorical values and balancing the dataset
to 7032 instances, emphasizing the telecom industry’s
dependence on customer feedback.
 Madhavi Kasa,in 2021 study compared Random Forest,
Decision Tree, and Na¨ıve Bayes algorithms, highlighting
the superior efficiency of the rough set-based model in
classifyingchurns from non-churns.

III. PROBLEM STATEMENT

In today’s highly competitive business world, customer Fig. 1. Flowchart of proposed system
retention is a critical factor for sustained growth and
profitability. Organizations face the challenge of high Here’s a detailed outline of Customer Churn Prediction
customer churn rates, which are affecting revenue and a system that involves the following steps:
decreasing the ability to achieve long-term success. To address  Collection of Bank Details Record.
this issue, we need an advanced customer churn prediction

IJISRT24MAR1246 www.ijisrt.com 2879

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

 Data Preprocessing.  k-Nearest Neighbour

 Classification model building. K-NN is a lazy learner while K-Means is an eager
 Splitting data into training data and test data. learner. An eager learner has a model fitting that means a
 Providing training data to the Classification model. training step but a lazy learner does not have a training
 Providing test data to the Evaluation model. phase.
 Prediction of Churner And non-Churner.
 SVM
A. Algorithms SVMs are particularly effective in solving classification
In this project, we are using methods for model problems, and they work by finding the optimal hyperplane
building such as classification algorithms which can detect that best separates data into different classes.
the accuracy of given input data, and from that, we can
compare which Machine learning algorithms are best for  XG Boost
implementing. In this project, our methodology involves the It is known for its exceptional predictive performance,
use of classification algorithms to build a predictive model speed, and versatility in handling various types of machine
that can accurately detect potential customer churn. We will learning tasks, particularly regression and classification.
collect and preprocess relevant customer data, perform
exploratory data analysis to gain insights into customer B. Methodology
behavior, select important features, and train a machine- The methodology for customer churn prediction involves
learning model to predict churn. We will evaluate the a systematic approach to build and deploy predictive models
model’s performance using various metrics and use clustering that can identify customers likely to churn.
algorithms to segment customers based on churn probability
or behaviour. Additionally, we will implement an early Here’s a step-by-step methodology for customer churn
warning system to alert stakeholders of potential churn risks. prediction:
To train the model we are using logistic regression, naive
Bayes, Decision tree, Random forest, k-neighbour, SVM, and  Data Collection and Preprocessing:
XG Boost classification algorithms.  Data Gathering: Collecting relevant customer data from
various sources
 Logistic Regression  Data Cleaning: Removing duplicates, handling missing
Logistic Regression is a statistical method used for values, and addressing data inconsistencies to ensure data
binary classification, which means it’s primarily used to quality.
predict one of two possible outcomes based on a set of input  Feature Engineering: Creating new features or
variables. In logistic regression, the dependent variable is transforming existing ones to extract relevant information
categorical, representing a binary outcome (e.g., yes/no, 1/0, that can improve the predictive model’s performance.
true/false).
 Exploratory Data Analysis (EDA):
 Naive Bayes Visualizing and exploring the data to get knowledge into
Naive Bayes classifier is a simple probabilistic machine customer behaviour,patterns, and potential churn drivers.
learning algorithm that is based on Baye’s theorem. it is
particularly useful for classification tasks where we can  Feature Selection
predict the categorical data. Removing irrelevant or highlycorrelated features to
avoid overfitting.
 Decision Tree
A decision tree algorithm is a popular machine learning  Model Selection:
technique used for both classifications. It is a tree-like model Choosing appropriate machine learningalgorithms for
that makes decisions or predictions by recursively splitting the churn prediction.
data into subsets based on the most significant features.
 Model Training
 Random Forest Splitting the dataset into training and testing sets for
Random Forest is an ensemble machine learning model training and evaluation.
algorithm that works by constructing a multitude of decision
trees during training and combining their predictions to make  Model Evaluation
more accurate and robust predictions. Random Forest is Assessing the model’s performance using metrics like
known for its high predictive accuracy. It excels at capturing accuracy, precision, recall, F1 score, and AUC-ROC on the
complex relationships within the data, which is often test set.
necessary when predicting churn.

IJISRT24MAR1246 www.ijisrt.com 2880

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

 Customer Segmentation  True Negatives (TN):The number of correctly predicted

Utilizing clustering algorithms (e.g., k-means, non-churn cases.
hierarchical clustering) to segment customers based on churn  False Positives (FP):The number of non-churn cases
probability or behaviour. predicted as churn.
 False Negatives(FN):The number of churn cases
 Early Warning System predictedas nonchurn.
With the help of the threshold churn rate, we can alert
businesses so they can avoid loss.  Accuracy (ACC)

 Retention Strategies and Recommendations (TP + TN )/(TP + TN + FP + FN )

By applying different strategies And Providing (1)
actionable recommendations on how to improve customer
engagementand satisfaction Measures the overall correct predictions.

VI. IMPLEMENTATION SETUP  Recall

Implementation Setup for Customer Churn Prediction  Google Sheets(For CSV)

Model involves the following : Google Sheets allows easy import and management of
CSV data, enabling collaborative data preprocessing and
A. Details about Input to Systems analysis for customer churn prediction project, accessible
In customer churn prediction projects Input data plays an from any device with internet connection.
important role in training and testing the predictive model.
The quality and accuracy of Input data have a high impact  Hardwares
on the output of the predictive model. In our Project, we will Windows 10 / Windows 11 OS
use a banking dataset for training our prediction model by
using various machine learning algorithms This input data VII. RESULT
includes the following points:
The classification models were implemented with python
 Customer Demographics 3.12.1, Churn modelling in bank dataset on laptop with intel
Personal information about the customer, such as Name, core i5 processor.
age, gender, and geography.
Table 1 Performance Metrics for Classifiers
 Customer Account Information Parameters Logistic Regression Naive Bayes Random Forest
Details about the customer’s accounts, CustomerId, Accuracy 0.8 0.791 0.876
Tenure. Recall 0.065 0.085 0.502
ROC 0.519 0.522 0.733
 Financial History
Information related to the customer’s credit score,
TP/(TP + FN )
HasCrCard, and Estimated salary.
(2)
 Customer Interaction Measure the proportion of actual churn cases that were
Data about the customer’s interactions with the bank correctly predicted.
such as Is, NumOfProducts, ActiveMember.
 ROC Curve (Receiver Operating Characteristic)
 Transaction History Graph of the true positive rate against the false positive
Records of the customer’s past transactions such as
rate.
Balance, when the customer is Exited.
 AUC-ROC (Area under the ROC Curve):
B. Performance Evaluation Parameters
Compare the overall performance of the matrices based
Performance matrices are the set of matrices that help
on Churn and non-churn cases.
to assess the effectiveness of the predictive model. In our
project, we will use the following matrices.  Gini Coefficient
Measures the inequality in the model’s performance
 Confusion Matrix
based on correctly predicting churn and non-churn.
 True Positives (TP): The number of correctly predicted
churn cases.

IJISRT24MAR1246 www.ijisrt.com 2881

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

 Entropy VIII. CONCLUSION

Entropy can be used to evaluate the impurity of a set of
customer data with respect to churn rate. Through the implementation of various classifiers
including Random Forest, KNN, XGBoost, Decision Tree etc.
C. Software and Hardware Setup for customer churn prediction, it was observed that Random
Forest gives higher accuracy, particularly after hyper
 Softwares parameter tuning. This indicates the efficacy of Random
Following Software required for the implementation of Forest in handling complex data and optimizing predictive
Customer Churn Prediction Model. performance, highlighting its suitability for customer churn
prediction tasks.
 Jupyter
Jupyter Notebook facilitates customer churn prediction ACKNOWLEDGMENT
projects by providing an interactive environment for data
exploration, visualization, and iterative model development, We would like to express our sincere gratitude to our
enhancing collaboration and documentation through code, supervisor, Prof. Sawant S.P., for his invaluable guidance,
visualizations, and explanatory text in a single, shareable patience, and support throughout this research. His insightful
document. feedback and constructive criticism have been instrumental
in shaping this work. we would also like to thank the
 Chrome participants who generously gave their time and shared their
Chrome can leverage Jupyter Notebook to access web- experiences for this study. Finally, we would like to thank
based data sources and seamlessly integrate with Python my family and friends for their unwavering support and
libraries for data analysis and model building, streamlining encouragement.
the process of customer churn prediction within a familiar
browser environment. REFERENCES

Table 2 Performance Metrics For Classifiers [1]. Assef Jafar Abdelrahim Kasem Ahmad* and Kadan
Parameters K-NN SVM XGBoost Aljoumaa. Customer churn prediction in telecom using
Accuracy 0.770 0.788 0.867 machine learning in big data platform. Journal Of Big
Recall 0.083 0.049 0.552 Data, 2019.
ROC 0.508 0.506 0.746 [2]. Siva Bala Krishnan Akshara Santharam. Survey on
customer churn prediction techniques IRJET, 2018.
[3]. Rushikesh Chavan Shradha Mandan Amol Chole, Trupti
Mate. Customers churn prediction using machine
learning and deep learning. JETIR, 2023.
[4]. Dr.K.Geetha. Customer churn prediction. Turkish
Journal of Computer and Mathematics Education, 2022.
[5]. Alaa Mostafa Abdelrahman Fahad Kamal Alsheref
Essam Abou el Kassem, Shereen Ali Hussein. Customer
churn prediction model and identifying features to
increase customer retention based on user generated
content. IJACSA, 2020.
[6]. Edemealem Desalegn Kingawa Tulu Tilahun Hailu.
Customer churn prediction using machine learning
techniques: the case of lion insurance. Asian Journal of
Fig. 2. Graph of Exited Vs Age Basic Science Research, 2022.
[7]. Prateek Anand Kiran Dhangar. A review on customer
The above graphs show us the churn rate with age. churn prediction using machine learning approach.
Visualizing where age is taken on the x-axis and exit istaken IJIERT, 2021.
on the y-axis. [8]. N.G.L. Prasanna R.Vindhya K.Sandhya Rani, Shaik
Thaslima and P. Srilakshmi. Analysis of customer churn
prediction in telecom industry using logistic regression.
IJIRCST, 2021.
[9]. Shreyas Rajesh Labhsetwar. Predictive analysis of
customer churn in telecom industry using supervised
learning. IJSC, 2020.

IJISRT24MAR1246 www.ijisrt.com 2882

Volume 9, Issue 3, March – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/IJISRT24MAR1246

[10]. Sena KASIM Levent Ç ALLI. Using machine learning

algorithms to analyze customer churn in the software as a
service (saas) industry. JESS, 2022.
[11]. Dr. S.B. Kishor Madhuri D. Gabhane, Dr. Aslam
Suriya. Churn prediction in telecommunication business
using cnn and ann. Journal of Positive School
Psychology, 2022. [12] Ming-Chang Qian Tong Ming
Zhao, Qingjun Zeng and Jiafu Su. A prediction model
of customer churn considering customer value: An
empirical research of telecom industry in china. Hindawi,
2021.
[12]. Mukta Sawant Shradhha Rajput Prof. Abhay Gaidhani
Mittal Patil, Mahima Rawal. Customer churn prediction.
IJCRT, 2021.

IJISRT24MAR1246 www.ijisrt.com 2883