European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024

Print ISSN: 2054-0957 (Print),

Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK

Real Time Credit Card Fraud Detection and Reporting

System Using Machine Learning
Jolaosho Ahmed Oluwatoyin1, Solomon Akinola2,
1
Department of Computer Science, Faculty of Natural and Applied Sciences,
Lead City University, Ibadan, Oyo State, Nigeria,
2
Department of Computer Science, Faculty of Natural and Applied Sciences, Lead City
University, Ibadan, Oyo State, Nigeria

doi: https://doi.org/10.37745/ejcsit.2013/vol12n43656 Published June 06, 2024

Oluwatoyin J.A.and Akinola S. (2024) Real Time Credit Card Fraud Detection and Reporting System Using Machine
Learning, European Journal of Computer Science and Information Technology, 12 (4),36-56

ABSTRACT: This study addresses the critical issue of real-time credit card fraud detection using
machine learning. The primary goal is to develop a model that promptly identifies fraudulent
transactions and alerts users. Two algorithms—Random Forest and Decision Tree Classifier were
used, alongside various sampling techniques to balance the dataset and enhance performance. Six
models were created, each with different accuracy levels in fraud detection. Key findings include
a higher incidence of fraud among individuals over 75 years, likely due to less familiarity with
modern transaction methods. Additionally, a majority of transactions involved females, indicating
a potential higher fraud risk in these transactions. The Random Forest -SMOTE [Hyperparameter
Tuned] model was the most effective, achieving a 97% accuracy rate, 95% F1 score, and 98%
precision rate. For practical application, this model was integrated with Twilio for real-time fraud
alerts, proving successful in sending timely, accurate notifications. The study highlights valuable
insights and a robust solution for real-time fraud detection and response. Regular performance
evaluations of the model are recommended to maintain its effectiveness against evolving fraud
patterns.

KEYWORDS—algorithm, credit card fraud, machine learning, real-time detection, twilo

integration

INTRODUCTION
The widespread adoption of online banking platforms, which facilitate various financial activities
such as cash transactions, bill payments, and online buying, has greatly enhanced the ease
experienced by a large number of persons in their everyday lives. Despite the considerable benefits
associated with conducting transactions online, customers are confronted with a substantial
36
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
challenge stemming from the occurrence of financial fraud. Fraud is the act of an unauthorised
individual or organisation bypassing the security mechanisms of a financial institution and
adopting the identity of a genuine customer in order to deceive and defraud the institution.
Financial fraud is a pervasive phenomenon that is steadily increasing and carries substantial
consequences inside the financial sector.
According to data from 2019, the worldwide population of individuals utilising credit cards
amounted to approximately 2.8 billion. Moreover, a minimum of 70% of these individuals are in
possession of at least one credit card (AlEmad, 2021). The United States experienced a notable
rise in credit card fraud cases, with a 44.7% increase observed between 2019 and 2020. The
number of reported incidents surged from 271,927 in 2019 to 393,207 in 2020. According to the
most recent study, the global losses resulting from credit card fraud in 2021 reached a total of
$32.34 billion, exhibiting an increase of nearly 14% compared to the losses experienced in the
previous year, which amounted to $28.43 billion (Mullen, 2023). According to Mullen (2023), The
aggregate losses due to fraudulent activities in the United States during the calendar year 2021
reached a sum of $11.91 billion, indicating a notable increase of 18% compared to the $10.09
billion reported in the year 2020. The occurrence of card not present fraud has exhibited a notable
rise of 81% in comparison to point of sale fraud. Credit card fraud is a pervasive problem in the
United States, as evidenced by the regular reporting of cases by banks, businesses, and cardholders.
Indeed, the nation is responsible for 38.6% of the global recorded losses incurred from credit card
fraud (Ashraf et al., 2022).

Between the months of January and September in the year 2020, financial services firms in Nigeria
incurred a financial loss amounting to ₦5.2 billion as a result of fraudulent operations (NIBSS,
2021; Anowu et al., 2021). The predominant portion of the financial loss was suffered within the
timeframe spanning from July to September 2020, wherein enterprises encountered losses
amounting to as much as ₦3.36 billion. During the corresponding period in 2020, there was a
notable escalation in the financial losses incurred as a result of fraudulent activities, exhibiting a
substantial jump of 510% in comparison to the ₦550 million lost in the preceding year of 2019
(NIBSS, 2021)..

Machine Learning is a specialised domain within the study of Artificial Intelligence that facilitates
the acquisition of knowledge and skills by computers through the process of learning from
experience, without the need for direct human intervention. The primary aim of this endeavour is
to forecast forthcoming results with a notable level of precision by employing various
computational models. According to Saravanan & Sujatha (2018) and Nasteski (2017) Machine
learning approaches can be classified into two primary categories: supervised learning and
unsupervised learning. Machine learning algorithms utilise past data to train models capable of
accurately predicting fraudulent transactions. It has become a widely used method for identifying
financial fraud because of its capacity to evaluate large volumes of data and uncover intricate
patterns that are challenging to detect using conventional techniques. Commonly employed

37
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
machine learning methods for fraud detection encompass logistic regression, decision trees,
random forests, support vector machines, and neural networks (Kumar et al., 2020; Hasan et al.,
2019).

This study aims to create a prototype model for real-time credit card fraud detection and
notification using machine learning algorithms. The objective is to enhance the accuracy and
efficiency of fraud detection and reporting. Two algorithms, namely Random Forest, and Decision
Tree Classifier, will be employed to identify instances of credit card fraud based on transaction
time and amount. Upon the identification of fraudulent activity, an automatic text message will be
promptly dispatched to notify the cardholder of the detected anomaly.

LITERATURE REVIEW

Financial Fraud
There are different categories of financial fraud in which credit or debit card fraud is the most
common. There are four types of Credit Card Fraud: Card not present, Skimming, Phishing and
Lost/Stolen Card (Popat & Chaudhary, 2018).
In Card not Present (CNP) fraud, fraudster attempt to mislead the system by dissembling to be
some other person (Popat & Chaudhary, 2018). Mail and the web are major routes for fraud against
merchants who sell and ship merchandise, and affects legitimate mail order and web merchants.
In Skimming, they are obtaining personal data regarding someone else's credit card utilized in an
otherwise normal transaction (Ileberi et al., 2020). There is a tiny device (skimmer) which is used
to swipe and store huge amount of victim’s information. In phishing, Scammers might use a range
of schemes to lure users into giving them their card info through tricks corresponding to websites
simulation to be of a bank or payment system. When card is steal or lost, there are chances for a
thief that he make unauthorized transaction before cardholder block the card (Popat & Chaudhary,
2018).
According to Shakya (2018) , the total amount lost due to fraudulent activities in 2018 has reached
$22.8 billion. This figure represents a 4% increase compared to the losses incurred in 2015.
Furthermore, it is anticipated that the losses will continue to rise significantly in the forthcoming
years. In the United Kingdom, the total amount of unauthorised fraud losses incurred by payment
cards, remote banking, and checks in the year 2022 amounted to £726.9 million (Connect, 2023).
This figure represents a marginal decrease of less than one percent when compared to the same
losses recorded in the year 2021.Remote purchase fraud, which involves the illicit utilisation of
pilfered card information to make online, telephone, or mail order purchases, continues to account
for the largest portion of financial losses, amounting to £395.7 million. However, it is worth noting
that this figure has decreased compared to the previous year. The incidence of fraud on lost and
stolen cards experienced a notable 30% increase, amounting to £100.2 million. Additionally,

38
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
instances of card identity theft, characterised by the unauthorised opening or takeover of a card
account under another individual's name, nearly doubled, reaching £51.7 million (Connect, 2023).
Anowu et al. (2021) reported that financial services firms in Nigeria experienced a loss of ₦5.2
billion due to fraudulent activities over the period spanning from January to September 2020. The
majority of this financial loss was incurred during the period from July to September 2020, with
enterprises seeing a total loss of up to ₦3.36 billion (Paul, 2021). There was a significant increase
of 510% in the amount lost to scammers, rising from ₦550 million in the corresponding period of
2019. Based on the 2021 fraud report published by the Nigerian Inter-Bank Settlement System
(NIBSS), there was a significant surge in fraudulent activities in Nigeria, with a notable increase
of 187% in the number of fraud attempts recorded between the years 2019 and 2020 (NIBSS
Insight, 2021). In 2020, the primary sources of fraudulent activities were identified as the internet,
accounting for 47% of reported cases, followed by mobile devices at 36%. ATM terminals
accounted for 9% of reported fraud incidents, while point-of-sale (POS) terminals constituted 7%
of the total cases (NIBSS, 2022)
Based on the data, the total amount of fraudulent activity amounted to $5.6 billion in 2012.
However, in 2018, the magnitude of fraud increased to $9.1 billion, representing nearly 40% of
the whole loss (Shakya, 2018). Specifically, 70% of the aforementioned fraudulent activities are
to Card-Not-Present (CNP) transactions, which encompass fraudulent incidents occurring online
or via telephone communication. Counterfeit transactions account for 20% of the total, while the
remaining 10% of cases involve losses incurred as a result of lost or stolen cards (Shakya, 2018).
The strategies for addressing fraud can be classified into two main categories: prevention, which
entails measures aimed at averting fraud at its source, and detection, which involves actions
conducted after to the occurrence of fraudulent activity Nevertheless, some authors have
undertaken efforts to combat this type of financial fraud.

Sudha & Akila (2021) proposed a Credit Card Fraud Detection system based on Operational &
Transaction features using Support Vector Machine (SVM) and Random Forest (RF) classifiers.
In this system, in the first phase, the operational features of users are extracted, and then a random
forest classifier is used to classify the features into benign and suspected. In the second phase, the
transaction features of users are extracted from the user records, and then the M-class SVM
classifier is applied to classify the features into benign and suspected. The performance of the
system is evaluated in terms of standard measures , accuracy, recall, and F-1 score. By results, it
was shown that both RF and SVM classifiers achieveprecision a higher detection rate with good
accuracy. Abdulghani et al (2021) proposed some of the classification ML algorithms such as
Logistic regression(LR), Linear Discriminant Analysis (LDA), and Naïve Bayes(NB), additionally,
the boosting algorithm XGBoost to create models capable of detecting fraud. The dataset from
Kaggle. The authors used performance metrics such as accuracy, precision, f1, recall, AUC
confusion matrix to evaluate the models' performance. The XGBoost model presented the best
results compared to other models

39
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Lavanya (2023) detected credit card fraud using random forest classifier in comparison with
logistic regression. Novel random forest classifier algorithm with sample size =09 and novel
logiswhen compared to logistic regression algorithm efficiency(95.34%). The statistical
significance difference(two-tailed) is 0.001(p<0.05).Their result showed that the Random tic
regression algorithm with sample size =09 were evaluated many times to predict the efficiency
percentage.The G power taken as 0.8 and a =0.05.Random forest classifier has been
efficiency(97.12%) forest classifier with 97.12% has better accuracy than 95.34% Logistic
regression in Credit card fraud detection. There is a statistical 2-tailed significant difference in
accuracy for algorithms is 0.02(p<0.05) by independent t-test. Abdulghani (2022) proposed a
model to detect fraudulent transactions based on some methods of machine learning such like
Logistic regression (LR), NaiveBayes (NB), as well as the Linear Discriminant Analysis(LDA),
in addition to XGBoost algorithm. The techniques used to generate the models of the suggested
system, they were trained and evaluated on the basis of two different datasets. Where the first
dataset was the European Cardholders, and the second dataset was the Turkish dataset provided
by the Yapi Kredi Company. These datasets suffer from a big imbalance problem. Therefore, the
authors used SMOTE to fix this issue. The system models' effectiveness was measured employing
a variety of metrics, specifically the confusion matrix, accuracy, F1score, recall, precision as well
as AUC. Also, the fuzzy membership function was adopted to the dataset in order to raise the
system's efficiency. The final results showed the high efficiency of XGBoost.

Krishna (2022) identified the frauds committed using a payment card such as credit cards, debit
cards, and also an experiment is performed to find the best suitable algorithm among Random
forest and Logistic Regression. To stop the fraud detections using Random forest (N=10) and
Logistic regression (N=10) with supervised learning that gives insights from the previous data.
The precision of the random forest is 76.29% compared with Logistic regression with accuracy of
74.65% with statistical significance value p=0.03 (p<0.05) using Independent sample t test.
Conclusion: This results proved that Random forest was significantly better for Fraud detection
than Logistic regression within the study’s limits. Ogundokun et al. (2021) proposed a credit card
fraud discovery scheme to detect fraud. The ML techniques employed are Decision Tree (DT) and
K-Nearest Neighbor (KNN) ML classification techniques. The performance outcomes of the two
ML classification techniques are evaluated depending on accuracy, precision, specificity, recall,
f1-score, and false-positive rate (FPR). The area under the ROC curve (AUC) of the receiver
operating characteristics (ROC) curve was similarly drawn built on the confusion matrix for both
classifiers. The two classification techniques were evaluated and compared using the performance
metrics mentioned earlier and it was demonstrated that the KNN technique outperformed that of
the DT with a greater ROC curve value of 91% for KNN and 86% for DT. It was concluded that
KNN is considered a better ML classification technique that can be employed to discover credit
card fraudulent activities.

Abakarim et al. (2018) proposed a live credit card fraud detection system based on a deep neural
network technology. The proposed model is based on an auto-encoder and it permits to classify,
40
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
in real-time, credit card transactions as legitimate or fraudulent. To test the effectiveness of our
model, four different binary classification models are used as a comparison. The Benchmark shows
promising results for our proposed model than existing solutions in terms of accuracy, recall and
precision. Also, Kumar et al. (2021) proposed a real time fraud detection system based on service
oriented architecture (SOA) to analyze the fraudulent activities in credit card transactions. The
architecture is designed on the basis of Apache Kafka tool, which is used for real time streaming
of transactional data to detect the fraudulent transactions. This process considers different services
which form the backbone of SOA. Further, five different machine learning classifiers namely
support vector machine (SVM), multilayer perceptron (MLP), random forest regressor,
autoencoder and isolation forest have been considered to identify the fraudulent activities instantly
with the help of SOA based real time architecture.

THEORETICAL BACKGROUND

Below, we explain the classification algorithms that were used in this paper.

Decision Tree
A decision tree is a form of supervised machine learning that may be used for classification or
regression tasks. It is particularly useful when the data can be separated based on specific
parameters, and it provides a visual depiction of the potential answers. All decisions were
contingent upon a multitude of conditions. The process begins at the root node and then diverges
into several solutions, resembling a tree structure. The tree originates from the root, subsequently
extending branches and progressively increasing in size. The primary concept is constructing a
tree T based on our collection of observations S. If every element in set S is a member of class C,
then the node is considered a leaf node and is assigned a label. Otherwise, the algorithm proceeds
to the next characteristic that provides the most relevant information and constructs sub-trees until
the desired objective is achieved. Firstly, it is necessary to establish the most informative attribute
by utilising the entropy approach, which quantifies the uniformity of the collected data (Mohamed
et al., 2020).

Entropy S = − ∑ P(x)log 2 P(x)

1

Also,the information gain that measures the relative change in entropy with respect to the
independent attribute is given as:

|Sv |
Gain(S, A) = Entropy(S) − ∑v∈A x Entropy(Sv )
S
2

41
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Random Forest Algorithm

Random forest is a supervised ensemble method that uses a collection of numerous decision trees
to make predictions (Shaik & Srinivasan, 2019). Random Forest is a classifier consisting of a set
of tree-structured classifiers with identically distributed independent random vectors and each tree
casting a unit vote at input x for the most popular class (Reis et al., 2018). A random vector that is
independent of the previous random vectors of the same distribution is generated and a tree is
generated using the training test , an upper bound is extracted for Random Forests to get the
generalization error in terms of two parameters Exactitude and interdependence of individual
classifiers (Yigin et al., 2020)

Figure 1: Random Forest Flow Chart (Arista, 2022).

The RF algorithm is very efficient, as it handles datasets that contain continuous variables, as well
as categorical variables robustly. An RF classifier contains subsets of various tree classifiers
{𝒉(𝒙, 𝚯𝒌 ), 𝒌 = 𝟏, 𝟐, … } where the 𝚯𝒌 are independently and identically distributed random
vectors, with each tree being able to specify the modal class at input 𝒙 (Tretiak et al., 2023). The
performance index, which solely approximates the confidence interval (CI) of the RF model is
given as

𝑚𝑔(𝑥, 𝑦) = 𝑎𝑣𝑘 𝐼(ℎ𝑘 (𝑥, Θ𝑘 ) = 𝑦) − max 𝑎𝑣𝑘 𝐼(ℎ𝑘 (𝑥, Θ𝑘 ) = 𝑗)

𝑗≠𝑦
3

where 𝐼(. ) denotes an indicator function, and 𝑎𝑣(. ), the average value. It is observed that as the
margin increases, the confidence level also increases. The generalisation error becomes

𝑃𝐸 ∗ = 𝑃𝑥,𝑦 (𝑚𝑔(𝑥, 𝑦) < 0),

4

where 𝑃(. ) denotes probability. With an increase in trees for all sequences Θ𝑘 , 𝑃𝐸 ∗ converges to

42
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
𝑃𝑥,𝑦 (𝑃Θ (ℎ(𝑥, Θ) = 𝑦) − max 𝑃Θ (ℎ(𝑥, Θ) = 𝑗 < 0)
𝑗≠𝑦
5

Convergence of this generalisation error proves that the RF model does not overfit as more trees
are introduced. The upper bound for the generalisation error is given as

̅ (1−𝑠2 )
𝜌
𝑃𝐸 ∗ ≤ ,
𝑠2
6

where 𝜌̅ is the average correlation value, s is the strength of each tree in the model. An increased
strength of individual trees and a low correlation between them produces more accurate prediction
results

MATERIALS AND METHODS

The system is designed such that, credit card comprehensive data will be collected from a
simulated credit card transaction dataset containing legitimate and fraud transactions and
preprocessed to normalize the data. Data sampling (Oversampling, undersampling and SMOTE)
was done to cater for imbalanced data that may affect the performance of the model. Relevant
features and variables were also identified that may influence the credit card fraud detection rate
through exploratory data analysis and domain expertise. Machine learning models was built using
two algorithms (Random Forest and Decision Tree) to detect credit card fraud, thus sending a real
time notification of any fraudulent activity. The performance of the system was evaluated and
interpreted using the evaluation metrics (Accuracy, F1 Score, Recall and Precision).
Logistic Regression,
Transaction Classification
Random Forest,
Dataset (Yes/No)
Decision Tree

Genuine Fraudulent
Transactions Transactions

Generate
Report

Figure 2: Design Framework

Data splitting: The dataset was divided into two distinct subsets: a Training set of 70% and a Test
set of 30% of the data. The training set was for resampling, hyperparameter tuning, and training
the model and the test set was used to test the performance of the trained model.
Data Resampling: This is use to handle the imbalanced dataset. If an imbalanced dataset is used,
the model built tends to be biased towards the legitimate transactions, and hence, it results in the
43
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
poor performance of model when tested in an unseen data. To tackle this problem, three resampling
techniques was used such as random undersampling, random oversampling, SMOTE. The
resampling technique was implemented on the training data separately to make it balanced.

Algorithm Models: Random Forest and Decision Tree Classifier models was designed and trained
using the preprocessed dataset. Three (3) sampling techniques was used (over sampling,
undersampling and SMOTE) in order to balance the dataset.

Evaluation Metrics
Evaluation metrics is used for evaluating the performance of the model depending on the nature
of the problem (whether it is a regression or classification). This thesis is limited to evaluation
metrics related to the classification problem.

Confusion Matrix : It is the most commonly used evaluation metrics in predictive analysis mainly
because it is very easy to understand and it can be used to compute other essential metrics such as
accuracy, recall, precision, etc (Shakya, 2018). It is an NxN matrix that describes the overall
performance of a model when used on some dataset, where N is the number of class labels in the
classification problem (Tiwari, 2022). A confusion matrix is composed of statistics such as True
Positive (TP), True Negative (TN), False Positive (FP) and False Negative (FN) which are
calculated using the combination of actual and predicted values (Anand, Velu & Whig, 2022).
True Positive (TP) is a case where the actual value was positive (e.g., fraud) and the predicted
value is also positive.
False Positive (FP) is a case where the actual value was negative (e.g., normal) but the predicted
value is positive.
True Negative (TN) is a case where the actual value was negative (e.g., normal) and the predicted
value is also negative.
False Negative (FN) is a case where the actual value was positive (e.g., fraud) but the predicted
value is negative.

Recall: Recall, also known as sensitivity, is the fraction of true positives to the actual positive
cases, which is shown in equation 7. In simple terms, recall is how many of true positives were
found (recalled) out of all the true positive cases.
𝑇𝑃
𝐑𝐞𝐜𝐚𝐥𝐥 =
𝑇𝑃+𝐹𝑁
7

Precision: As shown in equation , precision is the fraction of true positives over the true positives
and false positives. In simple terms, precision is how many of the found cases were true positives.
𝑇𝑃
𝐏𝐫𝐞𝐜𝐢𝐬𝐢𝐨𝐧 = 𝑇𝑃+𝐹𝑃
8

44
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
F1 Score: F1 Score also called F score or F-measure is the harmonic mean of the recall and
precision (Hand, Christen & Kirielle, 2021). Its value ranges from 0 to 1, where 0 is considered
worst, and 1 is considered best. It can be calculated as follows.
2∗ (Precision ∗ Recall)
𝐅𝟏 = (Precision + Recall)
9
Real Time Reporting: Twilio was be used in generating a real time text message if fraud is
detected. Twilio is a web application programming interface (API) that can be used to add
communications such as phone calling, messaging into Python applications.

Data Description

We utilised the open source data available at kaggle.com, namely the fraudTrain.csv file (Credit
Card Transactions Fraud Detection Dataset, 2020). This file contains a simulated credit card
transaction dataset that includes both valid and fraudulent transactions. It spans the period from
1st Jan 2019 to 31st Dec 2020 and includes the credit cards of 1000 customers who made
transactions with 800 merchants. We imported required packages and plotted the distribution of
each variable trans_date_trans_time', 'cc_num', 'merchant', 'category', 'amt', 'first', 'last', 'gender',
'street', 'city', 'state', 'zip', 'lat', 'long', 'city_pop', 'job', 'dob', 'trans_num', 'unix_time', 'merch_lat',
'merch_long', 'is_fraud. Also, analyzed the transaction patterns of the customers as shown in
figure 3.

The numeric values in the dataset was looked into, to get a sense of what it was like. The output
of the describe function on the dataset is shown in Figure 4.3.

Figure 3: Snapshot of the Numerical Values in the Dataset (Describe Function Analysis)

45
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Feature Engineering
The number of fraud per month was identified and analyzed as shown in figure 4.4

Figure 4: Number of Fraud Per Month

From the above visualizations it can be deduced that,

i. Moreover, according to the 'trans_year_month' data, most transactions have happened in the
January, February, October, December months of 2019 and in the May month of 2020.
ii. Also, it can be noted that the number of fraud customers and the number of fraud transaction
have increased in the time of December, which again is the holiday season.

Gender fraud distribution was grouped and plotted andage-fraud distribution was created as shown
in Figures 5 and 6

Figure 5: Distribution of Fraud Over Gender Chart

46
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
From the above visualization it can be observed that women contribute the most to the amount of
the transaction frequencies.Although women do participate in fraud, the amount of women
involved in fraud with respect to the number of transactions involving women is 0.52% whereas
the same for men is about 0.64%.It can be concluded that women are involved in most of the
transactions and hence, they be more prone to frauds.

Figure 6: Age-Fraud Distribution Chart

From the plots it can be observed that the most number of transactions in the datset have been
done by the people in the 30-45 age group. Also, the 46-60 age group has done significant number
of transactions.With respect to the total number of transaction made by a particular age group the
people in the >75 age group are the most affected, wherein, about 1% transaction made by these
people have been fraudulent.

3.3 Correlation Analysis of the Numeric Features

Figure 7: A Correlation Matrix of the Numeric Features

47
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
The main focus is on the correlation of the is_fraud column (indicating fraud) with other
columns.The highest correlation value is approximately 0.22, indicating a positive correlation
between the amt (transaction amount) and the likelihood of fraud (is_fraud). This suggests that
higher transaction amounts might be associated with a slightly higher likelihood of fraud.There
are some strong positive correlations, as indicated by coefficients close to 1:Strong positive
correlation between unix_time (correlation of approximately 0.999). Strong positive correlation
between long and merch_long, as well as between lat and merch_lat. This is likely due to the
relationship between transaction locations and merchant locations.Most other correlations are
close to 0, indicating weak or no linear relationships between the variables.For example, there are
weak correlations between transaction locations (lat, long) and the target variable (is_fraud).

RESULTS AND DISCUSSIONS

Machine Learning Models and Performance Evaluation

Two algorithms was used on the processed dataset and three(3)sampling techniques
(Oversampling, Undersampling and smote) to balance the dataset was used. Six (6) different
models was created and used for the prediction of credit card fraud. The features 'zip', 'lat', 'long',
'city_pop', 'unix_time', 'merch_lat', and 'merch_long' have been assumed to provide no significant
information in the model-building phase.

Decision Tree Classifier

Confusion Matrix for Decision Tree Classifier

Figure 7: Decision Tree Confusion Matrix

From the figure 7 the model correctly identified 1871 instances as "isfraud", correctly identified
385794 instances as "notfraud", the model made 924 false positive predictions and the model made
414 false negative predictions.

48
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Random Forest

Figure 8: Confusion Matrix for Random Forest Classifier

From figure 8, implies that the model correctly identified 1954 instances as "isfraud".The model
correctly identified 372657 instances as "notfraud", the model made 14061 false positive
predictions and the model made 331 false negative predictions.

Table 1: Summary of all Classification Model

Model Name Trainin Testing Accurac F1 Precisio Recall
g Score Score y Score n
1 Decision Tree - 0.99862 0.95426 0.998026 0.99794 0.89037 0.75711
imbalance class 2 3 6 6 2
2 Decision Tree -Random 0.98743 0.95426 0.954263 0.95426 0.95137 0.95729
Under Sampling 8 3 3 0 5
3 Decision Tree -Random 0.99129 0.94610 0.946100 0.94610 0.93726 0.95524
Over Sampling 7 0 0 5 3
4 Decision Tree - 0.99494 0.94021 0.940217 0.94021 0.95054 0.93010
SMOTE[Hyperparamet 9 7 7 9 8
er Tuned]
5 Random Forest - 0.99999 0.99825 0.998257 0.99814 0.95421 0.73873
imbalance class 7 7 7 1 1
6 Random Forest - 1.00000 0.96580 0.965808 0.96580 0.95511 0.95373
Random 0 8 3 7 7
Under Sampling
7 Random Forest - 1.00000 0.95497 0.954978 0.95496 0.96714 0.94117
Random Over Sampling 0 8 5 8 6
8 Random Forest - 1.00000 0.95380 0.973804 0.95380 0.95511 0.95340
SMOTE 0 4 5 7 5
[Hyperparameter
Tuned]
49
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
We used underfitting and overfitting, especially the imbalance dataset, to build and benchmark the
model against those built with the SMOTE Hyperparameter dataset because
i. To Evaluate Model Robustness: By comparing models trained on imbalanced datasets (which
might lead to underfitting or overfitting) with those trained on SMOTE-enhanced datasets, we
evaluated the robustness of the model. This comparison helps in understanding how well the
model can generalize from the training data to unseen data.
ii. To Demonstrate the Effectiveness of SMOTE: By benchmarking against models trained on
imbalanced datasets, we demonstrated the effectiveness of SMOTE in improving model
performance, especially in terms of metrics like precision, recall, and F1-score.
iii. To Highlight the Challenge of Imbalanced Datasets: Imbalanced datasets are a common
challenge in machine learning, where one class significantly outnumbers the other. By
showing how models perform poorly on imbalanced datasets, we highlighted the need for
techniques like SMOTE.
iv. To Understand Model Behavior: Observing how a model overfits or underfits in the context
of an imbalanced dataset can provide insights into its learning behavior. It can help in tuning
the model or choosing a more appropriate model or technique for the given problem.

From table 1 a total of eight distinct models have been generated. Among the eight (8) models
constructed, determine which model is the best based on Accuracy, F1 Score, Precision, and
Recall, Random Forest - SMOTE [Hyperparameter Tuned] the highest accuracy (0.973804), the
highest F1 Score (0.953805), which is a balanced measure of precision and recall, the highest
precision (0.977211), indicating that when it predicted "isfraud," it was usually correct, highest
recall (0.953737), indicating that it was effective at capturing a large proportion of "isfraud" cases.
Therefore, Random Forest Classifier - SMOTE sampling with hyperparameter tuning is the
optimal model for predicting credit card fraud in this work exhibiting an accuracy rate of 97.4%

Real Time Notification

Twilio was used in generating a real time text message if fraud is detected. The steps below was
used to integrate Twilio the fraud detection system:
Detect Fraudulent Activity: The machine learning model (RF hyperparameter Tuned) will
continuously monitor transactions or activities for signs of fraud.
Trigger an Alert: When the fraud detection system identifies a potentially fraudulent transaction
or activity, it them an alert to notify the appropriate personnel or system.
Integrate Twilio: Integrate Twilio's API into the model. Twilio provides APIs for sending SMS
messages programmatically.
Craft the Message: The default message was defined to notify when a fraud is detected. For this
research, the crafted message is “Fraudulent transaction detected on your credit card! Please
contact your bank immediately”.

50
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Testing and Validation: The model was tested with integration with Twilio to ensure that messages
are sent accurately and in a timely manner. The snippet code for the testing and validation is given
below

Figure 9: Sample of Simulated SMS Alert

DISCUSSION
Various features of the data set have been analyzed and several insights have been obtained. The
'trans_date_trans_time' feature has been broken down into several components like 'Age', 'day of
the week', 'month' in order to facilitate our analysis. These features have been thoroughly analyzed.
It has been found that old age people above 75 years are more susseptible to frauds. This is because,
fraudsters might try to take advantage of their lack of knowledge about the constantly changing
ways of how transactions are made.

The 'Female' gender people have been observed to do much of the transaction according to the
dataset. Hence, transactions involving might be much prone to fraud. Also, by analyzing several
demographic variables like city, state, zip etc it has been found that, several places like 'DE' state
has 100% fraud rate and about 50 zip codes and 70 cities have 100% fraud rate. There might be
some ill practices happening at the ground level at these places since all the transactions happening
there are shown as fraudulent.

Two (2) different algorithms have been implemented upon the processed dataset. Three sampling
techniques in order to balance the dataset have also been implemented. The algorithms have also
been implemented upon the dataset before balancing the dataset for demonstration purposes.
Hence, six (6) different models have been created the results of which have been summarized
above. Out of the 6 models that have been built, the Random Forest Classifier built using the
SMOTE sampling technique after hyper parameter tuning has provided the most preferable model

51
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
with a accuracy of 0.97, f1 score of 0.95 and precision of 0.98. Hence, it can be said the the Radom
Forest Classifier - SMOTE [Hyperparameter Tunned] sampling is the best model. Further, the
model (Radom Forest Classifier - SMOTE [Hyperparameter Tunned]) was tested with integration
with Twilio and the message was sent accurately and in a timely manner.
The result of this study has higher accuracy compared to a result of Thennakoon et al. (2019) that
reported that the machine learning models (LR, NB, LR and SVM) used in detecting credit card
fraud captured the four fraud patterns (Risky MCC, Unknown web address, ISOResponse Code,
Transaction above 100$)with an accuracy rate of 74%, 83%, 72% and 91% accuracy rates
respectively. This findings also corroborates the result in More et al. (2021) that reported an
accuracy of 97% when compared with Decision Tree and Naive Bayes Technique for credit card
fraud detection using supervised learning approach.

CONCLUSION AND RECOMMENDATION

The analysis and model development for fraud detection have provided valuable insights and a
robust solution for real time identifying and responding to fraudulent activities. The combination
of data analysis, feature engineering, and the selection of an effective machine learning model has
resulted in a powerful fraud detection system. The "Random Forest Classifier - SMOTE
[Hyperparameter Tuned]" model, along with Twilio integration, provides a comprehensive
solution for detecting and responding to fraud in a timely and accurate manner. This system is
well-equipped to safeguard against fraudulent transactions and protect both the business and its
customers. Based on the findings from this study, the following recommendations were made:
i. Continuous use and deploy the "Random Forest Classifier - SMOTE [Hyperparameter
Tuned]" model as the primary fraud detection model. It has demonstrated strong performance
across multiple metrics, including accuracy, F1 score, and precision
ii. Maintain the integration with Twilio for real-time SMS alerts. This is to ensure that the alerting
system is continuously monitored and tested to confirm its reliability and responsiveness.
iii. Perform periodic evaluations of the fraud detection model's performance to ensure its
effectiveness in detecting evolving fraud patterns. Revisit and retrain the model as necessary
to adapt to changing circumstances.
iv. Continue to explore and engineer features that could enhance the model's predictive
capabilities by considering additional data sources or variables that may provide valuable
insights into fraud detection.
v. Implement a robust monitoring and reporting system to track the model's performance in a
real-world environment. Ensure that alerts are acted upon promptly and that false
positives/negatives are reviewed and addressed

52
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Contributions and Limitation
The work demonstrates the application of various machine learning models and data sampling
techniques for fraud detection. It provides insights into which techniques are most effective in
handling imbalanced datasets and improving the accuracy of fraud detection systems. It also
showcases the importance of feature engineering and in-depth data analysis in understanding
transaction patterns and identifying potential fraud indicators. Breaking down transaction
timestamps and analyzing demographic variables offer valuable insights. The integration of the
fraud detection model with Twilio for real-time SMS alerts demonstrates a practical and proactive
approach to fraud prevention. It showcases how technology can be leveraged to respond swiftly to
potential fraudulent activities. The emphasis on continual improvement underscores the dynamic
nature of fraud detection. It encourages organizations to adapt and evolve their strategies to stay
ahead of fraudsters.

However, the research is limited by the availability of real life and historical data credit fraud data.
The data used was an open source dataset for training and evaluating the machine learning
algorithms. Also, another limitations is the potential challenges associated with adapting the
model to evolving fraud patterns and the need for continuous model updates. Another limitation
of the research is the potential for false positives, where legitimate transactions are incorrectly
flagged as fraudulent.
Suggestion for Further Studies
More research can be done to explore the application of deep learning models, such as
convolutional neural networks (CNNs) and recurrent neural networks (RNNs), for fraud detection.
Deep learning models can capture intricate patterns in data and may improve fraud detection
accuracy. Further research can be done by combining multiple data sources, such as transaction
data, user behavior data, and network logs, to create a multi-modal fraud detection system. Fusion
techniques can improve accuracy by leveraging complementary information. More work can be
done to investigate the challenges and opportunities of fraud detection in real-time data streams,
where transactions and events are continuously generated. Implement stream processing
techniques for timely detection. Investigation of advanced feature engineering techniques can be
explored, including natural language processing (NLP) for analyzing transaction descriptions and
customer reviews, to extract valuable information for fraud detection.

REFERENCES

Abakarim, Y., Lahby, M. and Attioui, A., 2018, October. An efficient real time model for credit
card fraud detection based on deep learning. In Proceedings of the 12th international
conference on intelligent systems: theories and applications (pp. 1-7).
Abdulghani, A., 2022. Employing machine learning techniques and fuzzy membership for
detecting fraud transactions in credit card (Master's thesis, Altınbaş Üniversitesi/Lisansüstü
Eğitim Enstitüsü).
53
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Abdulghani, A.Q., Uçan, O.N. and Alheeti, K.M.A., 2021, December. Credit card fraud detection
using XGBoost algorithm. In 2021 14th International Conference on Developments in
eSystems Engineering (DeSE) (pp. 487-492). IEEE.
AlEmad, M., 2022. Credit Card Fraud Detection Using Machine Learning.
Anand, M., Velu, A., & Whig, P. (2022). Prediction of loan behaviour with machine learning
models for secure banking. Journal of Computer Science and Engineering (JCSE), 3(1), 1-
13.
Anowu, D.N., Nyor, T., Agbi, S.E., Nelson, A.I. and Saliu, A.N., 2021. Financial forensic analysis
and fraud deterrence in listed deposit money banks in NIGERIA. Gusau Journal of
Accounting and Finance, 2(4), pp.18-18.
Arista, A., 2022. Comparison Decision Tree and Logistic Regression Machine Learning
Classification Algorithms to determine Covid-19. Sinkron: jurnal dan penelitian teknik
informatika, 7(1), pp.59-65.
Ashraf, M., Abourezka, M. A., & Maghraby, F. A. (2022). A Comparative Analysis of Credit Card
Fraud Detection Using Machine Learning and Deep Learning Techniques. In Digital
Transformation Technology: Proceedings of ITAF 2020 (pp. 267-282). Springer Singapore.
Connect, C. (2023, May 12). Over £1.2bn Lost to Fraud in 2022. Credit Connect. Retrieved
November 30, 2023, from https://www.credit-connect.co.uk/news/consumer-
lending/fraud/over-1-2bn-lost-to-fraud-in-2022/
Credit Card Transactions Fraud Detection Dataset. (2020). Kaggle.
https://www.kaggle.com/datasets/kartik2112/fraud-detection?select=fraudTrain.csv
Fraud in Nigerian Financial Services” 2021 https://nibss-
plc.com.ng/media/PDFs/post/NIBSS%20Insights%20Fraud.pdf
Hand, D. J., Christen, P., & Kirielle, N. (2021). F*: an interpretable transformation of the F-
measure. Machine Learning, 110(3), 451-456.
Hasan, M., Islam, M.M., Zarif, M.I.I. and Hashem, M.M.A., 2019. Attack and anomaly detection
in IoT sensors in IoT sites using machine learning approaches. Internet of Things, 7,
p.100059.
Ileberi, E., Sun, Y. and Wang, Z., 2022. A machine learning based credit card fraud detection using
the GA algorithm for feature selection. Journal of Big Data, 9(1), pp.1-17.
Krishna, M.V. and Praveenchandar, J., 2022, October. Comparative analysis of credit card fraud
detection using logistic regression with random forest towards an increase in accuracy of
prediction. In 2022 International Conference on Edge Computing and Applications
(ICECAA) (pp. 1097-1101). IEEE.
Kumar, A., Prusti, D., Purusottam, I.S. and Rath, S.K., 2021, July. Real time SOA based credit
card fraud detection system using machine learning techniques. In 2021 12th International
Conference on Computing Communication and Networking Technologies (ICCCNT) (pp.
1-6). IEEE.
Lavanya, K., 2023. A Comparison of Logistic Regression Classifier and Random Forest Classifier
for the Accurate Classification of Credit Card Fraudulent Transactions. Journal of Survey
in Fisheries Sciences, 10(1S), pp.2008-2017.
54
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Mohamed, S., Ashraf, R., Ghanem, A., Sakr, M. and Mohamed, R., Supervised Machine Learning
Techniques: A Comparison.
More, R., Awati, C., Shirgave, S., Deshmukh, R., & Patil, S. (2021). Credit card fraud detection
using supervised learning approach. International journal of scientific & technology
research, 9(10), 216-219.
Mullen, C. (2023). Card industry’s fraud-fighting efforts pay off: Nilson Report. Payments Dive.
https://www.paymentsdive.com/news/card-industry-fraud-fighting-efforts-pay-off-nilson-
report-credit-debit/639675/
NIBSS Insight, “fraud in nigerian financial services” 2021 https://nibss-
plc.com.ng/media/PDFs/post/NIBSS%20Insights%20Fraud.pdf
Ogundokun, R.O., Misra, S., Ogundokun, O.E., Oluranti, J. and Maskeliunas, R., 2021. Machine
learning classification based techniques for fraud discovery in credit card datasets. In
Applied Informatics: Fourth International Conference, ICAI 2021, Buenos Aires,
Argentina, October 28–30, 2021, Proceedings 4 (pp. 26-38). Springer International
Publishing.
Paul, E. (2021). In 2020, Nigeria Lost ₦5b to Fraud in 9 Months: What You Need to Watch Out
For. Techpoint Africa. Retrieved November 30, 2023, from
https://techpoint.africa/2021/02/22/nigeria-lost-5b-fraud-2020/
Popat, R.R. and Chaudhary, J., 2018, May. A survey on credit card fraud detection using machine
learning. In 2018 2nd international conference on trends in electronics and informatics
(ICOEI) (pp. 1120-1125). IEEE.
Reis, I., Baron, D. and Shahaf, S., 2018. Probabilistic random forest: A machine learning algorithm
for noisy data sets. The Astronomical Journal, 157(1), p.16.
Shaik, A.B. and Srinivasan, S., 2019. A brief survey on random forest ensembles in classification
model. In International Conference on Innovative Computing and Communications:
Proceedings of ICICC 2018, Volume 2 (pp. 253-260). Springer Singapore.
Shakya, R. (2018). Application of machine learning techniques in credit card fraud
detection (Doctoral dissertation, University of Nevada, Las Vegas).
Sudha, C. and Akila, D., 2021, January. Credit card fraud detection system based on operational
& transaction features using svm and random forest classifiers. In 2021 2nd International
Conference on Computation, Automation and Knowledge Management (ICCAKM) (pp.
133-138). IEEE.
Thennakoon, A., Bhagyani, C., Premadasa, S., Mihiranga, S., & Kuruwitaarachchi, N. (2019,
January). Real-time credit card fraud detection using machine learning. In 2019 9th
International Conference on Cloud Computing, Data Science & Engineering
(Confluence) (pp. 488-493). IEEE
Tiwari, A. (2022). Supervised learning: from theory to applications. In Artificial intelligence and
machine learning for EDGE computing (pp. 23-32). Academic Press.
Tretiak, K., Schollmeyer, G. and Ferson, S., 2023. Neural network model for imprecise regression
with interval dependent variables. Neural Networks, 161, pp.550-564.

55
 European Journal of Computer Science and Information Technology, 12 (4),36-56, 2024
Print ISSN: 2054-0957 (Print),
Online ISSN: 2054-0965 (Online)
Website: https://www.eajournals.org/
Publication of the European Centre for Research Training and Development -UK
Yigin, B.O., Algin, O. and Saygili, G., 2020. Comparison of morphometric parameters in
prediction of hydrocephalus using random forests. Computers in Biology and Medicine,
116, p.103547.

56