International Journal on Recent and Innovation Trends in Computing and Communication

ISSN: 2321-8169 Volume: 11 Issue: 9

Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

Predicting Customer Churn in E-Commerce Using

Statistical and Machine Learning Methods
Vidya Rajasekaran1 , Latha Tamilselvan2
1,2 Department of Information Technology
1,2 B.S. Abdur Rahman Crescent Institute of Science and Technology
1,2 Chennai,India
2 latha.tamil@crescent.education

Abstract— This research work aims to develop prediction models and analytical insights to overcome customer churn issues through data-
driven approaches. The attrition rate of consumers in e-commerce is a significant issue requiring effective retention strategies. A novel
methodology is proposed comprising data preprocessing, using statistical analysis techniques developing the model and carrying out tailored
retention strategies. The model is used to identify crucial churn influencers and propose practical recommendations for enhancing consumer
retention. The significance of this work lies in its potential to allow e-commerce ventures with insights with the intention of price savings
strategies, enhanced revenue measures, and better consumer fulfillment. This research will influence the e-commerce business by facilitating
evidence-based methods for reducing customer turnover and increasing long-term customer value. The resulting accuracy of the proposed model
using Logistic Regression results in 87 percentage of accuracy which is a good metric to assess overall model performance. The Kaplan-Meier
curve is used to check the survival probability of consumers and identify consumers more likely to churn over time.

Keywords-Consumer churn; Statistical analysis; Kaplan-Meier curve; Consumer retention; Logistic regression; Prediction.

I. INTRODUCTION and machine-learning approaches. Businesses may get insights

that enable them to take prompt and data-driven steps to retain
Retaining consumers is vital for an online business to important consumers by examining the gradations of the
continue to be successful in today's vibrant and intensely customer's behavior, their purchase patterns, and engagement
competitive e-commerce environment. The e-commerce following indicators. A general overview of the problems with
business has grown dramatically over the last decade, and its client turnover faced by e-commerce firms and the importance
ongoing expansion has introduced new obstacles. Customers of resolving these issues are focused initially. The second
today have access to a plethora of alternatives often just a few section delves into the statistical and machine learning
clicks away, making it easier than ever for them to switch their techniques that may be used to forecast customer turnover in e-
loyalty to a new platform or brand. Consequently, to be commerce, providing a methodical overview of the several
competitive, e-commerce companies need to be quick to react, options open to e-commerce companies. The tactics and
flexible, and acutely aware of client preferences. E-commerce approaches that may be used in this model help e-commerce
businesses continue to struggle with the phenomena of businesses create practical, customer-based strategies to lower
customer churn when customers impede engaging with a brand churn, enhance customer pleasure, and eventually fuel
or platform [1]. It is difficult to fully grasp the intricate sustainable business development in an increasingly
mechanics of client attrition. Pricing, product quality, website competitive business [3].
usability, customer service, and even external variables like E-commerce enterprises have supreme access to
economic conditions and industry trends can all influence a information about their consumers' interactions, tastes, and
consumer's choice to churn. E-commerce businesses must thus habits in the age of data-driven decision-making. This
embrace cutting-edge analytical and predictive techniques to profusion of data has the potential to be a potent tool in the
acquire insights. Understanding the variables that contribute to fight against client turnover if properly utilized. Organizations
customer churn and implementing practical ways to reduce it may get useful insights from this data and forecast when a
are crucial business imperatives for e-commerce companies as client may be about to churn, enabling timely and specialized
the digital market persists to enlarge and diversify by actions. This is done by using statistical and machine learning
developing applications [2]. In addition to affecting an e- methodologies. The objective of this work is to provide e-
commerce company's profits, customer churn also influences commerce specialists, data analysts, and business executives
the revenue of the overall business. Businesses started with the skills and resources they need to accurately forecast
increasingly relying on statistical and machine learning and deal with consumer attrition. E-commerce companies can
techniques to predict and decrease customer attrition rates to predict client performance and acclimatize necessary marketing
solve this problem. These methods provide e-commerce by leveraging the power of statistical and machine learning
businesses the capability to proactively spot consumers in approaches.
danger of leaving the e-commerce business and apply specific
and relevant client retention efforts.
To foresee consumer attrition in the context of e-commerce,
this proposed research investigates the relevance of statistical

3968
IJRITCC | September 2023, Available @ http://www.ijritcc.org
International Journal on Recent and Innovation Trends in Computing and Communication
ISSN: 2321-8169 Volume: 11 Issue: 9
Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

II. LITERATURE SURVEY the e-commerce industry. Spatial analysis is commonly used to
The rise of the industrial agglomeration e-commerce evaluate geographical or location-based data and patterns, and
business has been strongly influenced by the quick it may be used to identify spatial trends or dependencies linked
development of technology and the economy, increasing to customer turnover in this context. Contrarily, machine
customer shopping experiences and removing logistical learning techniques are useful for processing and analyzing
obstacles [4]. This development potential is supported by enormous volumes of data, making it possible to spot patterns,
contemporary information technology and networks, with trends, and prediction models.
multimedia and industrial cluster marketing increasing E-commerce enterprises must be able to forecast client
relevance. Virtual e-commerce platforms in industrial clusters turnover to improve customer retention and marketing tactics
haven't, however, received much research. This study examines [11]. The model presented in this research uses support vector
virtual e-commerce models in diverse industrial machine (SVM) prediction and k-means customer
agglomerations, discussing big data, industrial agglomeration, segmentation to detect customer loss in B2C e-commerce.
and their connection to e-commerce. It emphasizes the Customers are divided into three groups using the approach,
significant growth in domestic internet users over the past ten which also identifies important customer segments. The work
years and the significance of e-commerce for current and shows that k-means clustering enhances prediction accuracy
upcoming company endeavors. The study concludes that when comparing SVM and logistic regression for churn
various e-commerce models in industrial agglomerations must prediction. The accuracy of SVM is higher than that of logistic
be adjusted to various contexts since they are context-specific regression. The management of customer interactions can
and cannot be generalized. benefit from these insights for B2C e-commerce businesses.
This work [5] focuses on how crucial client retention is to This study explores several data analysis strategies,
corporate success, particularly in the cutthroat environment of techniques, and tools that may be used to examine consumer
today. Losing clients, also known as customer churn, is a major behavior data and find trends or patterns that could point to
problem for newly established businesses. The idea of impending churn [12]. These techniques could consist of data
employing a Stacking Classifier, an ensemble learning mining, machine learning techniques, and statistical analysis.
approach, to assess and forecast customer turnover in e- The study's goal is to investigate how these methods might be
commerce data is introduced in the article as a way to address used in the particular context of cargo and logistics, where
this. This classifier includes knowledge from four basic keeping clients is crucial to preserving the profitability and
learners: KNN, SVM, RF Classifiers, and Decision Trees. long-term viability of the company.
The research [6] focuses on dealing with customer churn, E-commerce businesses gather a lot of client information,
which is an important issue for large companies, especially in such as search history, buying trends, reviews, and comments.
the telecommunications industry. Using machine learning This data may be used using machine learning and data mining
methods on a big data platform, the authors created a churn tools to assess customer behavior and detect possible attrition
prediction model to address this problem. They increased issues [13]. A popular supervised learning technique in
prediction accuracy by utilizing Social Network Analysis machine learning, the support vector machine focuses on both
(SNA) elements as well. regression and classification issues in predictive analysis. The
Customer retention is a crucial issue since the approach for forecasting E-Commerce customer attrition
telecommunications business is both oversaturated and very presented in this research uses support vector machines in
competitive [7]. Data mining techniques and data science conjunction with a hybrid recommendation strategy. Empirical
technology provide useful tools for anticipating client turnover data show that employing the integrated forecasting model
and enhancing customer loyalty. The study intends to analyze significantly improves several parameters, including coverage
alternative methods and build data science models to categorize ratio, hit ratio, lift degree, and accuracy rate.
consumers based on their propensity to quit a telecom firm. Considering all the relevant research mechanisms a novel
This study shows how these models may forecast and lower mechanism is proposed to predict consumer churn in e-
customer turnover by identifying the major causes of churn and commerce business platforms.
improving services. Enhancing customer loyalty, lowering III. PROPOSED METHODOLOGY
attrition, and improving company results may all be achieved
by implementing customer churn prediction models. The initial process to start with is data preprocessing. The
Machine learning techniques for anticipating client attrition efficiency of the prediction techniques relies on the quality of
were compared and analyzed by the authors in [8]. The the data because raw data causes inaccuracies and
research assesses the effectiveness of several machine learning inconsistencies and impacts the accuracy of the results. The
algorithms in the context of customer attrition prediction. dataset is preprocessed by handling missing values through
The authors [9] explore the world of e-commerce, paying removal, identifying and removing duplicates and outliers,
close attention to client turnover prediction. The research finally, the data consistency is ensured. The target variable
results presented in the article can give e-commerce enterprises churn is set by finding the consumers who haven't made
insightful information that will help them better understand successful purchases in the last six months. Feature engineering
consumer behavior and foresee turnover. The creation of more is carried out to create new features and transform the existing
successful customer engagement programs, retention tactics, features to enhance the performance of the model. The features
and marketing campaigns that are specifically tailored to each involved in predicting the churn value are identified. The
customer can result from this. features include frequency, purchase history, monetary value,
The authors in [10] combine machine learning and and consumer demographic information like age, gender,
geographical analytic methods, which proposes a thorough location, and salary. Additional features like time-stamped
approach for researching and forecasting consumer attrition in

3969
IJRITCC | September 2023, Available @ http://www.ijritcc.org
International Journal on Recent and Innovation Trends in Computing and Communication
ISSN: 2321-8169 Volume: 11 Issue: 9
Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

data, successful transactions, average order value, and return

information are also considered.
The dataset is split in a ratio of 70:15:15. 70% of the data is
used in training the model, 15% is used in validation, and the
remaining 15% is used for testing the model. The dataset is
balanced using techniques like oversampling and under
sampling. Scaling is carried out to transform the data for good
model convergence.
Exploratory Data Analysis is conducted to gain insights
from the data and to identify and understand the significant
patterns, the structures, and their relationships. The basic
statistics are displayed for the dataset having numeric features
using the Python panda library to describe the function and the
results are stated in Fig.1.
Figure 3. Countplot representing the distribution of customer segments.

The Bi-variate analysis is a statistical analysis technique

used to examine the relationship between two variables and
their relationships. A heat map is generated to calculate and
visualize the correlation matrix and its influences on the churn.
Fig. 4 represents the correlation matrix and Fig. 5 denotes the
churn analysis. The churn analysis provides insights into the
proportion of customers churned.

1
Figure 1. Statistical Representation of the Operational Dataset.

Univariate analysis for the data distributions is represented

in Fig.2 . The histogram plot represents the total purchase Figure 4. Relationship analysis using Correlation Matrix.
amount and the distribution of purchase amounts.
A counterplot as shown in Fig. 3 is used to visualize the
distribution of categorical variables of the customer segments.
It displays the number of customers in each segment.

Figure 5. Countplot representing distribution of customer segments.

Fig.6 denotes demographic analysis representing the age

and Fig.7 represents the behavioral analysis through frequency
Figure 2. Histogram Representation of Total purchase amount and frequency distribution based on the customer churn analysis.
of purchase.

3970
IJRITCC | September 2023, Available @ http://www.ijritcc.org
International Journal on Recent and Innovation Trends in Computing and Communication
ISSN: 2321-8169 Volume: 11 Issue: 9
Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

A Statistical test is carried out to compare the

characteristics and behaviors of churned and non-churned
consumers. The significant differences can be calculated using
either the t-test, chi-square test, or ANOVA test. ANOVA
(Analysis of Variance) is applied to analyze the variation
between different groups and categories to explore the
statistically significant differences among them. The obtained
values are F-Statistic: 7.2470099773916745 and P-Value:
0.007123421087408745. The resulting interpretation states
that the analysis has found statistically significant evidence
that the mean total purchase amounts differ among the
consumers churned and not. The obtained p-value is less than
0.05 and it indicates that observed differences are statistically
significant. This insight is used for decision-making and
developing customer retention strategies.
A model is developed using Logistic Regression to predict
customer churn. The model estimates the probability of the
customer churn based on the features. Logistic Regression
algorithm is used in several prediction applications [14] [15]
[16] [17]. Logistic Regression is applied to binary
classification problems and here the binary classification task
of categorizing consumers as churned or not based on the
features . The model is used to estimate the probability of
customer churn. The Sigmoid function is applied to transform
the linear combination of features between a probability value
Figure 6. Age Distribution graph based on frequency and customer churn. of 0 and 1. The equation of the Logistic function is stated
below in (1).
1
𝑃(𝑋 = 1) = 1+𝑒 −(𝑎0 +𝑎1 𝑦1+𝑎2𝑦2+⋯+𝑎𝑛 𝑦𝑛) (1)

Here, P(X=1) is the probability of customer churn, y1, y2,

y3, and yn are the features, and a0, a1, a2 and an, denote the
logistic model coefficient values.
After the proposed model is trained and evaluated it is
applied to make predictions on the unseen data. The
probability of customer churn is calculated for the given
consumer based on their feature values. Finally, a decision
threshold is set to segregate the consumers as either churned
Figure 7. Consumer Purchase Frequency Graph based on Churn analysis. “1” or non-churned “0”. We set a threshold score of 0.5 stating
that if the predicted probability is greater than 0.5 the
consumers are classified as churned and less than 0.5 as non-
The time series analysis is depicted in Fig. 8. The
churned. If the value is exactly 0.5, they are kept in further
timestamp is represented in date time format and is set as an
investigation analysis for further upcoming months.
index. It helps in calculating the last purchase and purchase
The strategies to enhance customer retention can be
frequency of the consumers.
practiced based on the derived predictions. A conclusion can
be made for consumers with a high probability of churning by
targeting marketing strategies like discounts and promotions.
Further, the consumer survival probabilities and analysis
can be implemented through Kaplan-Meier analysis
techniques. The technique is used in survival analysis
estimation and visualization of consumers over a certain
period. The resulting curves are used in analyzing the time at
which the consumer churn occurs. It is used to calculate the
probability of a consumer continuing with an e-commerce
platform over a particular time and find the probability of a
customer staying over different periods of months. Time–to–
event data is the information on the customer churn and the
duration the customer remains active on an e-commerce
Figure 8. Time series analysis representation graph.

3971
IJRITCC | September 2023, Available @ http://www.ijritcc.org
International Journal on Recent and Innovation Trends in Computing and Communication
ISSN: 2321-8169 Volume: 11 Issue: 9
Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

platform before churn. All the consumers will not experience overall instances. The calculated accuracy measure is 0.8783,
the churn factor in the stated time and some will be still active which states that the model correctly predicts 87.83 % of the
and not churned at the end of time. This Kaplan-Meier curve is instances in the dataset using the formula in (3).
used to handle this type of issue and make predictions for 𝑇𝑃+𝑇𝑁
consumers who have not churned. 𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = 𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁 (3)
The formula for the Kaplan-Meier curve estimator for a
specific time is represented in the below equation (2). In the A confusion matrix table is generated to evaluate the
̂ (𝑡) is the estimated survival probability performance of the model. There are four values required to
below equation, 𝐾
calculate the confusion matrix score. The TP (True Positives)
over a time t. 𝑡𝑥 denotes the customer churn time over some are churned consumers that are correctly predicted by the
time, 𝐶𝑥 is the customer churn observed over a certain time proposed model. The TN (True Negatives) are non-churned
and 𝑅𝑥 is non-churned consumers over time x. consumers that are also correctly predicted as negatives by the
𝐶 proposed model. The FP (False Positives) is negatives which
𝐾̂ (𝑡) = ∏ (1 − 𝑥 ) (2)
𝑅𝑥 are predicted as positives by the model and FN (False
𝑡𝑥 ≤𝑡 Negatives) is instances that are positive in reality but are
The time series analysis is further extended to find the predicted as negatives. The result states that there are 533 true
seasonality changes in consumer behavior and their positives correctly predicted as churned customers, 434 true
relationship with consumer churn. Hence a final decision negatives correctly predicted as not churned, 56 as false
support system is generated to predict he churn and provide positives that are incorrectly predicted as churned consumers,
relevant insights to enhance consumer retention in e- and 78 as false negatives are churned consumers incorrectly
commerce business. The results of statistical analysis and predicted as not churned as shown in Fig.9.
machine learning algorithms can be compared to derive a The precision, recall, and F1-score are calculated for both
holistic view of consumer churn. The Kaplan-Meier curve is churned (True) and non-churned consumers (False) using
examined to check how it differs in terms of survival formulas in (4) (5), and (6).
probability and identify which consumers are likely to churn 𝑇𝑃
over time. The results of logistic regression are used to 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = 𝑇𝑃+𝐹𝑃
(4)
determine the predictor's statistical significance for predicting
𝑇𝑃
customer churn. The resulting coefficient values indicate the 𝑅𝑒𝑐𝑎𝑙𝑙 = (5)
𝑇𝑃+𝐹𝑁
direction and the strength of the model. The features creating a
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ×𝑅𝑒𝑐𝑎𝑙𝑙
significant impact on the churn are identified and used in 𝐹1 𝑆𝑐𝑜𝑟𝑒 = 2 × (6)
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛+𝑅𝑒𝑐𝑎𝑙𝑙
developing targeted retention strategies. Hence a complete
and interpretable analysis of consumer churn can be developed For true the precision score is 89% which indicates that the
for e-commerce businesses to understand the survival instances predicted as true are certainly true and recall 85%
mechanism and retention dynamics of consumers. indicates true instances predicted correctly. The F1-score is
87% which is a balance between precision and recall. The non-
IV. RESULTS AND DISCUSSION churned consumers stated as false resulted in a precision score
of 87% indicating instances predicted as false are false, 90% of
The performance metric of logistic regression is evaluated
the recall score stating actual false correctly predicted, and 89%
using various metrics like accuracy, precision, recall, F1-score,
of the F1-score stating balance between the two metrics. The
and support score [18] [19] [20]. The results are depicted in
support metric indicates the number of instances in each class.
Fig. 9. The below-calculated metrics are used to determine how
The proposed model performs well with good overall accuracy
well the proposed model can distinguish between churned and
providing detailed insights into the model's performance for
retained customers.
both the churned and non-churned consumers and compared to
different models [21].
Then, statistical analysis for consumer survival is estimated
through Kaplan-Meier curves shown in Fig.10.

Figure 9. Performance analysis results of Logistic Regression model.

Accuracy is a commonly used performance evaluation

metric for evaluating classification models used in measuring
the proportion of the instances correctly classified from the Figure 10. Kaplan-Meier curves denoting consumer churn probability.

3972
IJRITCC | September 2023, Available @ http://www.ijritcc.org
International Journal on Recent and Innovation Trends in Computing and Communication
ISSN: 2321-8169 Volume: 11 Issue: 9
Article Received: 25 July 2023 Revised: 12 September 2023 Accepted: 30 September 2023
___________________________________________________________________________________________________________________

The x-axis of the above graph represents time intervals in [9] Lubis, A. R., Prayudani, S., Julham, Nugroho, O., Lase, Y. Y., &
Lubis, M. (2022). Comparison of model in predicting customer
weeks and the y-axis denotes the survival probability. Initially, churn based on users' habits on e-Commerce. 2022 5th
the curve starts at 1 representing consumers are active, and International Seminar on Research of Information Technology and
slowly decreases over time when the consumers are churned. IntelligentSystems(ISRITI). https://doi.org/10.1109/isriti56927.202
Further separate curves can also be generated to identify 2.10052834
[10] Matuszelański, K., & Kopczewska, K. (2022). Customer churn in
consumer churn and tailor specific retention mechanisms. retail e-Commerce business: Spatial and machine learning
approach. Journal of Theoretical and Applied Electronic Commerce
V. CONCLUSION Research, 17(1), 165-198. https://doi.org/10.3390/jtaer17010009
[11] Xiahou X, Harada Y. B2C E-Commerce Customer Churn
This proposed work scrutinizes the techniques and data- Prediction Based on K-Means and SVM. Journal of Theoretical
driven strategies used to predict client attrition in the e- and Applied Electronic Commerce Research. 2022; 17(2):458-475.
commerce region, offering insights into customer retention https://doi.org/10.3390/jtaer17020024
tactics. Predictive models and approaches must continually be [12] Sahinkaya, G. O., Erek, D., Yaman, H., & Aktas, M. S. (2021). On
enhanced due to the business's quick development and the data analysis workflow for predicting customer churn behavior
in cargo and logistics sectors: Case study. 2021 International
changing consumer preferences. A fascinating strategy is Conference on Electrical, Communication, and Computer
exposed by combining statistics and machine learning Engineering(ICECCE). https://doi.org/10.1109/icecce52056.2021.
techniques. The trade-offs between accuracy and 9514117
interpretability is highlighted by the comparative examination [13] Shobana J, Ch. Gangadhar, Rakesh Kumar Arora, P.N.
of the developed model. Understanding customer turnover Renjith, J. Bamini, Yugendra devidas Chincholkar,E-
commerce customer churn prevention using machine
depends greatly on the model's interpretability and explain learning-based business intelligence strategy, Measurement:
ability. E-commerce businesses can make use of interpretable Sensors,Volume 27,2023,100728,ISSN 2665-
models to obtain perceptive information and make data-driven 9174,https://doi.org/10.1016/j.measen.2023.100728.
decisions to facilitate and efficiently retain consumers. The [14] Baxani, R., & Edinburgh, M. (2022). Heart disease
proposed research recommends churn prevention techniques as prediction using machine learning algorithms logistic
regression, support vector machine and random forest
well as individualized marketing, targeted incentives, and classification techniques. SSRN Electronic
improved customer service to increase customer loyalty and Journal. https://doi.org/10.2139/ssrn.4151423
profitability. Future studies may examine real-time data [15] Edlitz, Y., & Segal, E. (2022). Author response: Prediction
analytics, sophisticated deep learning methods, and models that of type 2 diabetes mellitus onset using logistic regression-
based scorecards. https://doi.org/10.7554/elife.71862.sa2
adjust to shifting consumer behavior as the e-commerce [16] Harshini, P. S., Naresh, K., Pamulapati, S. R., &
industry changes. Lavanya, A. (2023). Diagnosis of liver diseases using
machine learning algorithms and their prediction using
REFERENCES logistic regression and ANN. 2023 3rd International
Conference on Intelligent Technologies
(CONIT). https://doi.org/10.1109/conit59222.2023.1020581
[1] Li, L. & Zhang, J. (2021). Research and Analysis of an 9
Enterprise E-Commerce Marketing System Under the Big [17] Kadam, E., Gupta, A., Jagtap, S., Dubey, I., & Tawde, G.
Data Environment. Journal of Organizational and End User (2023). Loan approval prediction system using logistic
Computing (JOEUC), 33(6), 1-19. regression and CIBIL score. 2023 4th International
http://doi.org/10.4018/JOEUC.20211101.oa15 Conference on Electronics and Sustainable Communication
[2] JWenli, Z., & Liang, H. (2023). Research on optimization Systems
design of fresh e-Commerce APP based on KANO (ICESC). https://doi.org/10.1109/icesc57686.2023.1019315
model. International Conference on Cyber Security, 0
Artificial Intelligence, and Digital Economy (CSAIDE [18] Manglani, R., & Bokhare, A. (2021). Logistic regression
2023). https://doi.org/10.1117/12.2681544 model for loan prediction: A machine learning
[3] Šaković Jovanović J, Vujadinović R, Mitreva E, Fragassa C, approach. 2021 Emerging Trends in Industry 4.0 (ETI
Vujović A. The Relationship between E-Commerce and 4.0). https://doi.org/10.1109/eti4.051663.2021.9619201
Firm Performance: The Mediating Role of Internet Sales [19] Rajasekaran, V., & Priyadarshini, R. (2021). An e-
Channels. Sustainability. 2020; 12(17):6993. Commerce prototype for predicting the product return
https://doi.org/10.3390/su12176993 phenomenon using optimization and regression
[4] Yue Hongqiang,Research on E-Commerce Data Standard techniques. Communications in Computer and Information
System in the Era of Digital Economy From the Perspective Science, 230-240. https://doi.org/10.1007/978-3-030-88244-
of Organizational Psychology,Frontiers in 0_22
Psychology,VOLUME=13,YEAR=2022,DOI=10.3389/fpsy [20] Reddy, D. J., Gunasekaran, M., & Sundari, K. S. (2022).
g.2022.900698, ISSN=1664-1078 undefined. 2022 International Conference on Cyber
[5] Awasthi, S. (2022). Customer churn prediction on e- Resilience
Commerce data using stacking (ICCR). https://doi.org/10.1109/iccr56254.2022.9995969
classifier. https://doi.org/10.36227/techrxiv.20291694 [21] T. Vafeiadis, K. I. Diamantaras, G. Sarigiannidis, and K. C.
[6] Demir, B., & Ergün, Ö. Ö. (2023). Customer churn prediction with Chatzisavvas, “A comparison of machine learning techniques for
machine learning methods in telecommunication customer churn prediction,” Simul. Model. Pract. Theory, vol. 55,
industry. https://doi.org/10.21203/rs.3.rs-3343217/v1 no. 10, pp. 1–9, 2015, doi: 10.1016/j.simpat.2015.03.003.
[7] Gopal, P., & MohdNawi, N. B. (2021). A survey on customer churn [22] Retail Case Study Analysis | Kaggle
prediction using machine learning and data mining techniques in e-
Commerce. 2021 IEEE Asia-Pacific Conference on Computer
Science and Data Engineering
(CSDE). https://doi.org/10.1109/csde53843.2021.9718460
[8] Karamollaoglu, H., Yucedag, I., & Dogru, I. A. (2021). Customer
churn prediction using machine learning methods: A comparative
analysis. 2021 6th International Conference on Computer Science
andEngineering(UBMK). https://doi.org/10.1109/ubmk52708.2021
.9558876

3973
IJRITCC | September 2023, Available @ http://www.ijritcc.org