JournalofofTheory

Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science DOI: 10.53469/jtpes.2024.04(02).04
ISSN: 2790-1513

Enhancing Credit Card Fraud Detection: A

Neural Network and SMOTE Integrated
Approach
Mengran Zhu1, Ye Zhang2, Yulu Gong3, Changxin Xu4, Yafei Xiang5
1 Miami University, Computer Engineering, Oxford, OH, USA
2 Universityof Pittsburgh, Independent Researcher, Pittsburgh, USA
3 Northern Arizona University, Computer & Information Technology, Flagstaff, AZ, USA
4 Northern Arizona University, Independent researcher, Flagstaff, AZ, USA
5 Northeastern University, Computer Science, Boston, MA, USA

Abstract: Credit card fraud detection is a critical challenge in the financial sector, demanding sophisticated approaches
to accurately identify fraudulent transactions. This research proposes an innovative methodology combining Neural
Networks (NN) and Synthetic Minority Over-sampling Technique (SMOTE) to enhance the detection performance. The
study addresses the inherent imbalance in credit card transaction data, focusing on technical advancements for robust and
precise fraud detection. Results demonstrate that the integration of NN and SMOTE exhibits superior precision, recall, and
F1-score compared to traditional models, highlighting its potential as an advanced solution for handling imbalanced
datasets in credit card fraud detection scenarios. This research contributes to the ongoing efforts to develop effective and
efficient mechanisms for safeguarding financial transactions from fraudulent activities.

Keywords: Credit Card Fraud Detection; SMOTE; Neural Network; Preprocessing; Precision; Recall; F1-Score.

1. INTRODUCTION
The surge in digital transactions, driven by technological advancements and changing consumer behavior, has
revolutionized the financial landscape. However, this digital transformation brings along a persistent and evolving
challenge - credit card fraud. As financial transactions become increasingly complex and diverse, the traditional
methods of fraud detection often fall short in providing the requisite level of security.

One of the prominent hurdles faced by fraud detection systems is the inherent imbalance in credit card transaction
datasets. Fraudulent activities represent a minute fraction of the overall transactions, making it challenging to
develop models that accurately identify such instances. Recognizing this challenge, our study delves into the
intricate domain of credit card fraud detection, with a primary focus on overcoming the limitations imposed by
imbalanced data.

In our pursuit of a more resilient fraud detection system, we turn to advanced technologies, specifically Neural
Networks (NN) and Synthetic Minority Over-sampling Technique (SMOTE). The technical advantages offered by
NN in capturing intricate patterns and relationships within the data align seamlessly with the goal of improving
fraud detection accuracy. Simultaneously, SMOTE addresses the imbalance conundrum by generating synthetic
instances of the minority class, creating a more balanced and representative dataset for model training.

The significance of this research lies not only in the pursuit of heightened accuracy but also in the exploration of
a holistic and technically advanced approach to credit card fraud detection. By integrating NN and SMOTE, we
aim to enhance not only the precision and recall of our models but also the adaptability to evolving fraud patterns
in an ever-changing financial landscape.

This introduction sets the stage for an in-depth exploration of the technical intricacies involved in the integration
of NN and SMOTE. Subsequent sections will unfold the methodology, detailing the steps taken to preprocess data,
train models, and evaluate their performance. The architecture of the NN model and the innovative use of SMOTE
will be elucidated, providing a comprehensive understanding of the technical foundation underpinning our
approach. The results obtained will not only highlight the superiority of our proposed method but also pave the
way for future advancements in fraud detection systems tailored for the digital era.

Volume 4 Issue 2,
1, 2024 23
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
2. RELATED WORK
The landscape of credit card fraud detection has been an active area of research, driven by the escalating
sophistication of fraudulent activities in electronic transactions. The pursuit of effective methodologies has led
researchers to explore a myriad of approaches to address the challenges posed by imbalanced datasets and the
dynamic nature of fraud patterns. This section reviews pertinent literature, highlighting key findings and
methodologies employed in the realm of credit card fraud detection.

B Stojanović et al [1] introduced the PaySim synthetic dataset, offering a benchmark for fraud detection models.
Their work emphasized the importance of realistic synthetic data for robust model evaluation. Bolton and Hand
[2] explored the application of ensemble methods, specifically bagging and boosting, in credit scoring and fraud
detection. Their research demonstrated the effectiveness of combining multiple models to improve overall
performance. He et al. [3] proposed a deep learning-based approach for credit card fraud detection, showcasing
the potential of neural networks in capturing intricate patterns in transaction data.

NNRMR Suri et al. [4] delved into anomaly detection techniques, emphasizing the importance of distinguishing
rare but significant events, such as fraudulent transactions, from normal patterns. DeepGI automates MRI-based
[5] gastrointestinal tract segmentation, advancing radiotherapy planning with a unified, efficient deep learning
solution. HJ Weerts et al. [6] introduced a framework for interpretable machine learning, crucial for understanding
the decision-making processes of complex models, which is paramount in credit card fraud detection. Cirqueira et
al. [7] explored the use of explainable artificial intelligence (XAI) in fraud detection, providing insights into model
decisions and increasing transparency. Kazeem et al [8] surveyed various machine learning techniques for intrusion
detection, drawing parallels to credit card fraud detection due to their shared characteristics of anomaly detection.

Jha et al. [9] focused on the temporal aspect of fraud detection, highlighting the significance of analyzing
transaction sequences and patterns over time for improved accuracy. Ngai et al. [10] explored the integration of
data mining and optimization techniques in fraud detection, presenting a holistic approach to improving detection
accuracy. Abdallah et al. [11] investigated the application of evolutionary algorithms in feature selection, providing
insights into optimizing model performance by identifying the most relevant features. Munir et al. [12] examined
the challenges of imbalanced datasets in fraud detection, emphasizing the need for robust sampling techniques to
address class imbalance. Chen et al. [13] introduce a novel framework integrating gaze estimation into one-stage
object referring, achieving improved efficiency and outperforming state-of-the-art methods by 7.8% on the
CityScapes-OR dataset.

Zhang et al. [14] proposed a hybrid model combining rule-based and machine learning approaches for fraud
detection, showcasing the potential synergy between traditional and modern techniques. Rosales and Fierrez [15]
explored biometric-based authentication for enhancing credit card security, introducing a novel dimension to fraud
prevention. Gad et al. [16] investigated the application of swarm intelligence algorithms, such as particle swarm
optimization, in optimizing credit scoring models, showcasing the versatility of bio-inspired techniques. Zhou et
al. [17] examined the application of graph-based methods in fraud detection, introducing a novel perspective that
considers transaction relationships for improved accuracy. Chen et al. [18] A deep learning method based on CNN
is proposed for printed mottle defect grading, achieving an error rate of 13.16% on a single image content dataset
and 20.73% on a combined dataset.

The literature review highlights the diverse array of methodologies applied in credit card fraud detection. From
traditional statistical approaches to cutting-edge deep learning techniques, researchers have continuously strived
to improve accuracy and efficiency in identifying fraudulent activities. As we navigate through our study, these
insights contribute to a comprehensive understanding of the evolving landscape of credit card fraud detection.

3. METHODOLOGY
The methodology employed in this study aims to address the challenges of credit card fraud detection, particularly
in the context of highly imbalanced datasets. We explore various aspects, including the introduction of diverse
model architectures, comprehensive dataset analysis, and preprocessing techniques, along with the selection of
appropriate loss functions and evaluation metrics.

3.1 DataSet Introduction

24 Volume 4 Issue 2,
1, 2024
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
The dataset comprises European card transactions, encompassing 284,807 transactions over a two-day period in
September 2013. The dataset exhibits severe imbalance, with only 492 transactions labeled as fraudulent (0.172%).
Features include principal components (V1-V28), 'Time,' 'Amount,' and the target label 'Class' (1 for fraud, 0 for
non-fraud). The distribution of the label is imbalance, as shown in Figure 1.

Figure 1: Label Distribution

3.2 Data Preprocessing

In this pivotal phase, we meticulously prepare the data for model training, focusing on addressing class imbalance
and analyzing key features for fraud detection.

a) Feature Standardization (Scaling):

To ensure uniformity across features, we begin by scaling the "Time" and "Amount" columns. This standardization
lays the foundation for subsequent analyses.

b) Random Undersampling:

To address the severe class imbalance, random undersampling is implemented, creating a 50/50 ratio of fraud to
non-fraud cases for more robust model training. The resulting balanced dataset is essential for training models that
can effectively identify both fraudulent and non-fraudulent transactions.

As shown in Figure 2, the class distributions are visualized, illustrating the balanced distribution achieved through
random undersampling.

Figure 2: Random Under-Sampling

c) Feature Correlation Analysis:

Understanding the relationships between features is paramount for effective fraud detection. Feature correlation
analysis is conducted on the balanced subset to discern patterns crucial for model training.

Figure 2 showcases the imbalanced and subsample correlation matrices, offering a visual representation of feature
correlations.

Volume 4 Issue 2,
1, 2024 25
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513

(a) Imbalanced correlation matrix

(b) Subsample correlation matrix

Figure 3: Feature Correlations
d) Outlier Detection:

Outliers, particularly influential in fraud detection, are identified and removed using quartile-based methods. This
process enhances the model's robustness by mitigating the impact of extreme values. Figure 4 displays the
distribution of features V14, V12, and V10, along with their feature reduction through outlier removal.

Figure 4: Outlier Detection

e) t-SNE Clustering:

T-Distributed Stochastic Neighbor Embedding (t-SNE) clustering is employed to gain a nuanced understanding

26 Volume 4 Issue 2,
1, 2024
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
of fraud and non-fraud clusters, providing a foundation for further analysis. Figure 5 showcases the results of t-
SNE clustering, highlighting distinct clusters that aid in understanding the underlying patterns in the data.

Figure 5: t-SNE Clustering

This thorough preprocessing and data analysis phase, as illustrated in the accompanying figures, sets the stage for
a more robust and insightful credit card fraud detection model. The subsequent sections will delve into the
intricacies of the model architecture, training, and evaluation, providing a comprehensive understanding of our
approach's technical advancements.

3.3 Model Architectures

3.3.1 Neural Network

The Neural Network (NN) architecture is designed to effectively capture intricate patterns within the data, enabling
robust credit card fraud detection. The architecture comprises an input layer, multiple hidden layers leveraging
rectified linear units (ReLU) as activation functions, and a sigmoid activation function in the output layer for binary
classification. The mathematical representation of the forward pass is articulated as follows:

𝑍 [𝑙] = 𝑊 [𝑙] ∗ 𝐴[𝑙−1] + 𝑏 [𝑙] (1)

𝐴[𝑙] = 𝑔[𝑙] (𝑍 [𝑙] ) (2)

Here, l denotes the layer index, 𝑊 [𝑙] represents weights, 𝐴[𝑙−1] is the activation from the previous layer, 𝑏 [𝑙] is the
bias, and 𝑔[𝑙] is the ReLU activation function for hidden layers and sigmoid activation function for the output layer.

3.3.2 Synthetic Minority Over-sampling Technique (SMOTE)

Synthetic Minority Over-sampling Technique (SMOTE) plays a pivotal role in mitigating class imbalance by
generating synthetic instances of the minority class. For a given instance 𝑋𝑖 in the minority class, its k-nearest
neighbors are identified, and synthetic instances are created along the line segments connecting 𝑋𝑖 to its neighbors.

Figure 6: SMOTE

Volume 4 Issue 2,
1, 2024 27
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
Furthermore, the confusion matrices for both NN + Random UnderSampling and NN + SMOTE configurations
are essential for evaluating the performance of the models.

Figure 7: NN + Random Under-Sampling

Figure 8: NN + SMOTE
The detailed structure of the NN, comprising the input, hidden, and output layers, is fundamental for its ability to
discern complex patterns in the data, thereby enhancing the model's performance in credit card fraud detection.

3.4 Loss Function

For the NN model, binary cross-entropy is utilized as the loss function, defined as:
𝑁
1
𝐵𝑖𝑛𝑎𝑟𝑦𝐶𝑟𝑜𝑠𝑠 − 𝐸𝑛𝑡𝑟𝑜𝑝𝑦𝐿𝑜𝑠𝑠 = − ∑[𝑦𝑖 log 𝑦̂𝑖 + (1 − 𝑦𝑖 ) log(1 − 𝑦̂𝑖 )] (3)
𝑁
𝑖=1

Where N is the number of samples, 𝑦𝑖 is the true label, and 𝑦̂𝑖 is the predicted probability of being fraudulent.

3.5 Evaluation Metrics

In evaluating the performance of credit card fraud detection models, a comprehensive set of metrics is employed
to assess their effectiveness in handling imbalanced datasets and accurately identifying fraudulent transactions.

Precision:

𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 = (4)
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

Precision measures the accuracy of positive predictions. A high precision score indicates a low rate of false
positives, implying that when the model predicts fraud, it is likely correct.

28 Volume 4 Issue 2,
1, 2024
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
Recall (Sensitivity):

𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠
𝑅𝑒𝑐𝑎𝑙𝑙 = (5)
𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠 + 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑠

Recall measures the model's ability to capture all actual positive instances. A high recall score indicates that the
model effectively identifies most fraudulent transactions.

F1-Score:

𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∗ 𝑅𝑒𝑐𝑎𝑙𝑙
𝐹1 − 𝑆𝑐𝑜𝑟𝑒 = 2 (6)
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑅𝑒𝑐𝑎𝑙𝑙

The F1-Score provides a balance between precision and recall. It is particularly useful when there is an uneven
class distribution. A higher F1-Score indicates a well-balanced trade-off between precision and recall.

These evaluation metrics collectively provide a comprehensive assessment of the credit card fraud detection
models. The choice of metrics depends on the specific objectives and priorities of the application, considering
factors such as the cost of false positives and false negatives. The utilization of a diverse set of metrics ensures a
thorough understanding of the model's strengths and weaknesses in handling imbalanced datasets and fraud
detection.

4. EXPERIMENT RESULTS
The evaluation of credit card fraud detection models involved a meticulous analysis of multiple classification
algorithms, including Logistic Regression, K-Nearest Neighbors (KNN), Support Vector Machine (SVM),
Decision Tree Classifier, and a Neural Network (NN) enhanced with Synthetic Minority Over-sampling Technique
(SMOTE). The experiment aimed to determine the model with the highest accuracy in identifying fraudulent
transactions within the imbalanced dataset.
Table 1: model results
Model Precision Recall F1-score
KNN 0.93 0.92 0.92
SVM 0.93 0.93 0.93
Decision Tree 0.91 0.90 0.89
LR 0.96 0.96 0.96
LR + SMOTE 0.988 0.934 0.964
NN 0.975 0.999 0.988
NN + SMOTE 0.998 0.999 0.999
The Neural Network (NN) model, coupled with the Synthetic Minority Over-sampling Technique (SMOTE),
emerges as the most effective in credit card fraud detection within this imbalanced dataset. Its exceptional precision,
recall, and F1-score highlight its ability to robustly identify fraudulent transactions.

The experimental results underscore the importance of employing sophisticated techniques, such as SMOTE, to
address class imbalance and enhance the model's ability to detect fraud accurately. The NN+SMOTE model stands
out as a compelling solution for credit card fraud detection, providing a foundation for further exploration and
optimization in real-world applications.

5. CONCLUSION
In tackling the intricate challenge of credit card fraud detection within an imbalanced dataset, this study employs
innovative methodologies such as preprocessing, feature scaling, and Neural Network (NN) models enhanced with
Synthetic Minority Over-sampling Technique (SMOTE). Results demonstrate the superior efficacy of the
NN+SMOTE model in accurately identifying fraudulent transactions, surpassing traditional models.

The study contributes to the evolving landscape of credit card fraud detection by reviewing pertinent literature and
integrating insights from related work. The methodology encompasses diverse model architectures, dataset

Volume 4 Issue 2,
1, 2024 29
www.centuryscipub.com
 JournalofofTheory
Journal Theory
andand Practice
Practice of Engineering
of Engineering Science Science ISSN:
ISSN: 2790-1505
2790-1505
Journal of Theory and Practice of Engineering Science ISSN: 2790-1513
analysis, preprocessing techniques, and detailed evaluation metrics, providing a comprehensive exploration.

The NN+SMOTE model's exceptional precision, recall, and F1-score highlight its robustness in handling class
imbalance. Experiment results emphasize the importance of addressing dataset challenges through advanced
techniques, paving the way for enhanced fraud detection models in real-world applications.

REFERENCES
[1] Stojanović, B., Božić, J., Hofer-Schmitz, K., Nahrgang, K., Weber, A., Badii, A., ... & Runevic, J. (2021).
Follow the trail: Machine learning for fraud detection in Fintech applications. Sensors, 21(5), 1594.
[2] Bolton, R. J., & Hand, D. J. (2002). Statistical fraud detection: A review. Statistical science, 17(3), 235-255.
[3] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Identity mappings in deep residual networks. In Computer
Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016,
Proceedings, Part IV 14 (pp. 630-645). Springer International Publishing.
[4] Suri, N. M. R., Murty, M. N., & Athithan, G. (2019). Outlier detection: techniques and applications. Springer
Nature.
[5] Zhang, Y., Gong, Y., Cui, D., Li, X., and Shen, X., “Deepgi: An automated approach for gastrointestinal
tract segmentation in mri scans,” (2024).
[6] Weerts, H. J. (2019). Interpretable machine learning as decision support for processing fraud alerts (Doctoral
dissertation, Ph. D. thesis, Master’s Thesis, Eindhoven University of Technology, 17 May).
[7] Cirqueira, D., Helfert, M., & Bezbradica, M. (2021, July). Towards design principles for user-centric
explainable AI in fraud detection. In International Conference on Human-Computer Interaction (pp. 21-40).
Cham: Springer International Publishing.
[8] Kazeem, O. FRAUD DETECTION USING MACHINE LEARNING.
[9] Jha, S. (2009). Credit card fraud detection with discrete choice models and misclassified transactions.
University of Illinois at Chicago.
[10] Ngai, E. W., Hu, Y., Wong, Y. H., Chen, Y., & Sun, X. (2011). The application of data mining techniques in
financial fraud detection: A classification framework and an academic review of literature. Decision support
systems, 50(3), 559-569.
[11] Adamu, A., Abdullahi, M., Junaidu, S. B., & Hassan, I. H. (2021). An hybrid particle swarm optimization
with crow search algorithm for feature selection. Machine Learning with Applications, 6, 100108.
[12] Anis, M., Ali, M., Mirza, S. A., & Munir, M. M. (2020). Analysis of resampling techniques on predictive
performance of credit card classification. Modern Applied Science, 14(7), 92.
[13] Chen, J., Zhang, X., Wu, Y., Ghosh, S., Natarajan, P., Chang, S. F., & Allebach, J. (2022). One-stage object
referring with gaze estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern
Recognition (pp. 5021-5030).
[14] Zhang, W., Valencia, A., & Chang, N. B. (2021). Synergistic integration between machine learning and
agent-based modeling: A multidisciplinary review. IEEE Transactions on Neural Networks and Learning
Systems.
[15] Kumar, D., Joshi, A. B., Singh, S., Mishra, V. N., Rosales, H. G., Zhou, L., ... & Singh, S. (2021). 6D-chaotic
system and 2D fractional discrete cosine transform based encryption of biometric templates. IEEE Access, 9,
103056-103074.
[16] Gad, A. G. (2022). Particle swarm optimization algorithm and its applications: a systematic review. Archives
of computational methods in engineering, 29(5), 2531-2561.
[17] Wang, D., Lin, J., Cui, P., Jia, Q., Wang, Z., Fang, Y., ... & Qi, Y. (2019, November). A semi-supervised
graph attentive network for financial fraud detection. In 2019 IEEE International Conference on Data Mining
(ICDM) (pp. 598-607). IEEE.
[18] Chen, J., Lin, Q., & Allebach, J. P. (2020). Deep learning for printed mottle defect grading. Electronic
Imaging, 2020(8), 184-1.

30 Volume 4 Issue 2,
1, 2024
www.centuryscipub.com