Enhancing Credit Card Fraud Detection Using DBSCAN-Augmented Disjunctive Voting Ensemble

Enhancing Credit Card Fraud Detection Using DBSCAN-Augmented Disjunctive Voting Ensemble | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Enhancing Credit Card Fraud Detection Using DBSCAN-Augmented Disjunctive Voting Ensemble Mahmoud A. Ghalwash, Samir Mohamed Abdelrazek, Nabila Hamid Eladawi, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7237183/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 13 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Credit card fraud detection remains a critical yet challenging task due to the extreme class imbalance inherent in transaction datasets, where fraudulent activities constitute only a small fraction of the total records. To address this imbalance and enhance the detection of rare fraud instances, this study proposes a novel hybrid framework that integrates density-based clustering for data augmentation with an ensemble classification strategy optimized for high recall. In the preprocessing stage, the proposed method utilizes Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to identify minority-class clusters and synthetically augment the fraud class. This step aims to preserve the intrinsic structure of fraudulent patterns while increasing their representation in the training set. Subsequently, an ensemble model comprising Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) classifiers is constructed. Final predictions are generated using a Disjunctive Voting Ensemble (DVE) strategy, wherein a transaction is classified as fraudulent if any of the base classifiers predicts it as such. This permissive voting mechanism prioritizes recall, thereby minimizing the risk of undetected fraudulent transactions. Extensive experiments conducted on three publicly available credit card fraud imbalanced datasets containing transaction records from European credit cardholders in 2023, providing a realistic fraud detection scenario. Computational efficiency in both training and testing is achieved using the Anaconda Navigator (Spider-Python 3.12) environment. The comparative analysis shows that the proposed DBSCAN-augmented DVE framework delivers notable improvements over traditional ensemble approaches and single-model baselines, particularly in recall and F1-score, while preserving consistently high precision. This approach proves to be both robust and interpretable, making it highly applicable to real-world fraud detection scenarios characterized by severe class imbalance. The results are especially compelling, achieving recall and F1-scores as high as 99.5% and 99.8% respectively, and consistently maintaining a perfect accuracy and precision of 100% across all imbalanced datasets. This study highlights the effectiveness of hybrid ensemble approaches in combating credit card fraud. The findings lay the groundwork for developing more resilient and adaptive fraud detection systems, which are crucial in countering the evolving tactics of fraudsters. The proposed model marks a significant advancement in securing financial transactions and mitigating risks in an increasingly digital economy. Physical sciences/Engineering Physical sciences/Mathematics and computing Credit card fraud detection hybrid ensemble approach ensemble learning machine learning data imbalance enhanced (augmented) dataset and disjunctive voting ensemble classification mechanism Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Introduction Fraudulent activities in the financial sector continue to rise. The use of credit and debit cards for online shopping has significantly increased due to the growth and optimistic outlook of e-commerce. However, this has also led to a heightened risk of credit and debit card fraud 1 . The Federal Trade Commission (FTC) reports that 2021 was a historic year for identity theft, highlighting that fraud losses increase more than 70 percent over 2020 to more than $5.8 billion 2 . The FTC emphasizes the urgent need for innovative solutions to protect both consumers and businesses from these threats. According to the United Kingdom Finance Annual Fraud Report 2022 , over £1.3 billion was stolen in 2021 through authorized and unauthorized criminal activities. Despite these challenges, the banking and finance sector prevented an additional £1.4 billion in unauthorized fraud, underscoring the effectiveness of existing measures 3 . In the United States, fraud cases have also surged. The FTC’s Consumer Sentinel Network 2022 Report recorded 2.4 million fraud complaints in 2022, with total losses reaching nearly $8.8 billion. Investment scams saw the most dramatic rise, with losses reaching nearly $3.8 billion in 2022, more than double the amount reported in 2021 2 . Fraud Detection and Ensemble Learning Techniques To address these increasing threats, a range of techniques is employed for credit card fraud detection, including statistical, machine learning, and deep learning approaches. Statistical methods such as regression, hypothesis testing, and clustering help identify anomalies in transaction patterns. Machine learning algorithms analyze historical data to detect fraud in real-time, while deep learning utilizes neural networks to uncover intricate patterns in large datasets, delivering high accuracy in fraud detection. A major challenge in credit card fraud detection is the imbalance in data, caused by the uneven distribution of fraudulent and non-fraudulent transactions. This imbalance can result in biased models and diminished effectiveness in identifying fraud. Research 4,5,6 has tackled this issue through approaches such as data balancing, oversampling, under-sampling, and the Synthetic Minority Oversampling Technique (SMOTE). However, a comprehensive evaluation of the effectiveness of these methods is still needed. Ensemble learning techniques, which integrate multiple models, play a crucial role in credit card fraud detection. Approaches such as bagging, boosting, and stacking are especially effective in managing data imbalance and improving predictive performance 7 . By capitalizing on the strengths of various base models, ensemble learning enhances accuracy while minimizing false positives and false negatives. This paper explores the challenges of credit card fraud detection and provides a review of the state-of-the-art techniques and evaluation criteria. The study aims to propose a framework for a hybrid ensemble of diverse machine learning models, benchmarking its performance against hybrid supervised/unsupervised models. The primary objective of a fraud detection model is to generate accurate alerts while minimizing false alarms and missed fraud cases. To achieve this, the study conducts a detailed comparative analysis between hybrid supervised/unsupervised models and ensemble models, utilizing various practical evaluation metrics to identify the superior approach for improving credit card fraud detection on transaction data. Objectives and Contributions This paper assesses the performance of a hybrid ensemble model that combines multiple algorithms while utilizing imbalanced datasets for credit card fraud detection. The key contributions include: Addressing Data Imbalance: Constructing a model tailored to mitigate the issue of disproportionate representation between fraudulent and legitimate transactions. Improving Computational Efficiency: Creating a hybrid ensemble framework optimized for handling intricate algorithms, sophisticated feature engineering, and a variety of base classifiers efficiently. Introducing a hybrid ensemble approach that leverages Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to refine data features. The ensemble integrates Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) classifiers. Final predictions are determined through a Disjunctive Voting Ensemble (DVE) mechanism. Performance Evaluation: Comparing the effectiveness of the proposed hybrid ensemble model against individual machine learning algorithms—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF)—as well as a traditional ensemble model that employs a voting strategy among these classifiers. The structure of this paper is as follows: Section 2 provides a review of related works, emphasizing machine learning and ensemble techniques for credit card fraud detection. Section 3 presents a detailed explanation of the proposed hybrid ensemble model. Section 4 states the metrics used to evaluate the performance of the models. The experimental setup is comprehensively described in Section 5. Section 6 provides a comprehensive analysis of the experimental results, including performance assessments and comparative evaluations of the proposed hybrid ensemble model. Finally, Section 7 concludes the paper with key findings and insights, along with discussions on future directions for enhancing fraud detection systems. Related Work This section explores existing literature on credit card fraud detection, with a focus on proposed systems and techniques, particularly Machine Learning and Ensemble Learning models. Machine Learning (ML) in Credit Card Fraud Detection Machine learning algorithms are essential for detecting credit card fraud, as they can analyze data, recognize intricate patterns, and predict fraudulent transactions. These algorithms fall into two main categories: supervised and unsupervised learning methods. Commonly used techniques for Credit Card Fraud Detection (CCFD) include Logistic Regression (LR), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Naive Bayes (NB), Decision Trees (DT), Random Forest (RF), and others. Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) classifiers (supervised and unsupervised), are among the most powerful machine learning models (used in fraud detection. SVM classifies data by finding the optimal hyperplane 8 , KNN classifies transactions based on the nearest neighbors 9 , RF aggregates decision trees to reduce overfitting 10 , and DBSCAN. These diverse approaches contribute to the robustness of fraud detection systems, offering effective ways to identify and prevent fraudulent transactions. Tanouz et al. 11 carried out an in-depth study on machine learning techniques for credit card fraud classification, with a particular emphasis on imbalanced datasets. Their findings demonstrated that Random Forest is a robust approach for fraud detection. However, the absence of a feature selection process constrained the models' performance. Raghavan et al. 12 investigated fraud detection by applying data mining techniques to three datasets from Australia (AU), Germany, and Europe (EU). Their study utilized algorithms such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest. Additionally, they developed two ensemble models: one integrating KNN, SVM, and Convolutional Neural Network (CNN), and another combining KNN, SVM, and Random Forest. The findings revealed that SVM outperformed the other algorithms. While their research provided valuable insights into the effectiveness of various algorithms and ensemble approaches for fraud detection, overall performance remained relatively low across all datasets. In 2022, Qaddoura et al. 13 examined the impact of different oversampling techniques, including SMOTE, ADASYN, borderline1, borderline2, and SVM-based oversampling algorithms. Their study revealed that applying oversampling methods can significantly improve model performance. In 2022, Sahithi et al. [1] presented a credit card fraud detection model based on a Weighted Average Ensemble, integrating LR, RF, KNN, Adaboost, and Bagging. Their study highlights the effectiveness of ensemble models in detecting credit card fraud within this critical domain. However, the limited discussion on the feature selection process affects the model's reproducibility. Ruttala et al. 14 conducted a comparative analysis of the Random Forest and AdaBoost algorithms for credit card fraud detection using an imbalanced dataset. Their findings revealed that Random Forest outperformed AdaBoost in terms of precision, recall, and F1-score. Tiwari et al. 15 conducted a comparative study of various credit card fraud detection techniques, assessing algorithms such as SVM, ANN, Bayesian Network, K-Nearest Neighbor (KNN), and Decision Trees. Using the KDD dataset from the KDD CUP 99 Intrusion Dataset, they found varying accuracy levels: SVM achieved 94.65%, ANN reached 99.71%, KNN attained 97.15%, and Decision Trees recorded 94.7%. While their analysis provided valuable insights into fraud detection methods, its effectiveness was limited by the dataset’s inability to fully represent real-world financial activities. Numerous studies have investigated methods to enhance fraud prevention and detection in credit card transactions using machine learning. Prasad Chowdary et al 16 . introduced an ensemble approach to improve CCFD. Their work emphasizes optimizing model parameters, improving performance metrics, and incorporating deep learning to minimize identification errors and false negatives. By combining multiple classifiers and conducting rigorous evaluations, their approach enhances the efficiency of CCFD systems. Sadgali et al. 17 aimed to determine the most effective techniques for detecting financial fraud. Their approach incorporated various methods, including Support Vector Machine (SVM). Notably, their study did not focus on a specific dataset for analysis. The results indicated that Naïve Bayes achieved the highest performance, with SVM closely following. However, the research was limited to insurance fraud detection. Saputra et al. 18 evaluated the performance of Decision Tree, Naïve Bayes, Random Forest, and Neural Network algorithms for fraud detection, utilizing SMOTE to mitigate dataset imbalance. The Kaggle dataset used in the study contained a low percentage of fraudulent transactions (0.093%). Based on confusion matrix analysis, the results showed that the Neural Network achieved the highest accuracy, followed by Random Forest. Additionally, SMOTE significantly improved the average F1-score, effectively handling the data imbalance. Forough et al. 19 developed an ensemble model that integrates deep recurrent neural networks with an innovative voting mechanism based on an artificial neural network to detect fraudulent activities. The model employs multiple recurrent networks, such as LSTM or GRU, as base classifiers, and combines their outputs using a feed-forward neural network (FFNN) for the voting process. The ensemble model, utilizing GRU, achieves its best performance with two base classifiers on both the European cards dataset and the Brazilian dataset. It outperforms the individual GRU model across all metrics and the baseline ensemble model in most metrics. Karthik et al. 20 proposed a novel credit card fraud detection model that combines ensemble learning techniques such as boosting and bagging. This hybrid classifier leverages the strengths of both methods, with Adaboost used for feature engineering of the behavioral feature space. The model's performance was assessed using the area under the precision-recall (AUPR) curve, showing moderate improvements, with results ranging from 58.03% to 69.97% on the Brazilian bank dataset and from 54.66% to 69.40% on the UCSD-FICO dataset. The Proposed Hybrid Ensemble Framework The Proposed Hybrid Ensemble Model The proposed hybrid ensemble model integrates Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) with feature engineering using DBSCAN. As an unsupervised clustering algorithm, DBSCAN detects anomalies and patterns within the dataset. Its outputs are incorporated as an additional feature, creating an augmented dataset that enhances the information available to the supervised classifiers. By applying the ensemble method to the augmented dataset, the model leverages the distinct advantages of its components: the robustness of Random Forest (RF), the simplicity of K-Nearest Neighbors (KNN), and the ability of Support Vector Machine (SVM) to handle complex decision boundaries. The integration of DBSCAN enhances the feature set by combining supervised learning with unsupervised clustering, which boosts both predictive accuracy and model robustness—particularly valuable in fraud detection scenarios. Furthermore, the use of disjunctive voting emphasizes recall, reducing the likelihood of overlooking fraudulent transactions and resulting in a more dependable system for fraud detection and other classification tasks. Methodology The Classification Phase : The proposed hybrid ensemble model integrates three key algorithms : دRandom Forest (RF): A supervised ensemble method that constructs multiple decision trees and aggregates their predictions to enhance accuracy. Support Vector Machine (SVM): A supervised learning model that identifies the optimal hyperplane to separate fraud and non-fraud transactions. K-Nearest Neighbors (KNN): A distance-based algorithm that classifies a sample based on the majority class among its nearest neighbors. Disjunctive Voting Mechanism In a Disjunctive (OR-based) voting scheme, a transaction is classified as fraudulent if any of the individual classifiers predicts it as fraud. This is advantageous in imbalanced settings where fraudulent transactions (minority class) are rare but critical to detect. It increases the sensitivity (recall) by ensuring that potential frauds are less likely to be missed. The predictions from RF, SVM, and KNN are combined using a disjunctive voting approach, where each algorithm's output contributes to the final classification decision. The Ensemble Disjunctive Voting Mechanism : The disjunctive voting mechanism among the three integrated algorithms ( RF, KNN, and SVM ) determines the final classification decision, as outlined in Table 1 : If RF, KNN, and SVM unanimously vote "non-fraud", the final decision of the ensemble model will be " non-fraud " . Otherwise, if any of the models vote “fraud” on a classification, this determines the final decision. Performance Metrics : Table 1 Ensemble Disjunctive Voting Decisions Models Classification RF Non-fraud fraud x x KNN Non-fraud x fraud x SVM Non-fraud x x fraud Final Classification Non-fraud fraud fraud fraud The following metrics are used to evaluate the performance of the models: Accuracy: Measures the overall performance of the model by calculating the percentage of correctly classified instances out of the total examined cases. Precision: Defined as the ratio of correctly predicted positive instances to the total predicted positives, precision assesses the accuracy of the model’s positive predictions. This metric is particularly important in scenarios where false positives carry significant consequences. Recall (Sensitivity): Represents the ratio of correctly predicted positive instances to the total actual positives, indicating the model’s ability to identify all relevant occurrences. High recall is crucial in applications were failing to detect positive cases could lead to serious consequences. F1 Score: A harmonic mean of precision and recall, offering a balanced evaluation of both metrics. It is particularly useful in cases with imbalanced class distributions , where accuracy alone may not provide a complete picture of model performance . Confusion Matrix: Provides a breakdown of model predictions into true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), enabling a detailed performance analysis. Confusion Matrix : (5) Figure 2. Illustration of the performance Metrics It is important to note that the F1 score is particularly useful when accuracy is misleading, especially in imbalanced datasets. It effectively balances precision and recall, addressing the following trade-offs: High precision, low recall → The model is too strict, resulting in fewer false positives but many false negatives. High recall, low precision → The model is too lenient, leading to more false positives but fewer false negatives. In credit card fraud detection, the goal is to achieve both high precision and recall to effectively prevent fraudulent transactions while maintaining system performance by accurately accepting legitimate ones. Experimental Setup : Case Study 1 Using The Original Imbalanced Dataset : The experiments are conducted on the original subset of 242,400 records that is split into three imbalanced subsets (40400, 80800, and 121200 records) as mentioned in section 3.2. The case study includes two Experiments. Train and test individual algorithms Each individual algorithm—RF, KNN, and SVM—is trained and tested on the three imbalanced datasets using an 80 − 20 split. Their effectiveness in detecting fraud (0) and non-fraud transactions (1) is measured using the performance metrics described in Section 4. Ensemble majority evaluation Additionally, a majority voting strategy is applied to the three ensemble supervised models (RF, KNN, SVM) trained on the imbalanced datasets. The ensemble model’s performance in identifying fraudulent (0) and non-fraudulent (1) transactions is also evaluated using the performance metrics outlined in Section 4. The performance of individual algorithms (RF, KNN, and SVM) is compared with the majority voting strategy of the three ensemble supervised models (RF, KNN, SVM) to evaluate their effectiveness in detecting both fraudulent and non-fraudulent credit card transactions. Case Study 2 Applying The Proposed Hybrid Ensemble Model : The experiments are carried out on the augmented dataset consisting of 242,400 records, which is divided into three imbalanced subsets (40,400, 80,800, and 121,200 records) as described in Section 3.4. The case study follows a two-step process: First, individual algorithms—RF, KNN, and SVM—are trained and tested on the three augmented imbalanced datasets using an 80 − 20 split. Their ability to classify fraud and non-fraud transactions is evaluated and incorporated into the disjunctive voting mechanism described in section 3.5.2.1. Second, the disjunctive voting mechanism for the proposed hybrid ensemble model, as described in Section 3.5.2.1, is applied to the classifications produced by individual algorithms (RF, KNN, and SVM) for final evaluation. The performance of this hybrid ensemble model in identifying fraudulent and non-fraudulent transactions is measured using the metrics specified in Section 4. The performance results of the proposed Hybrid Ensemble Model are documented and compared with those of other models to draw a conclusion. Implementation Setup : The models were implemented using the Spyder application for Python programming. The development process utilized Python along with libraries such as Scikit-learn, Pandas, and NumPy. Data preprocessing, model training, and evaluation were conducted within the IPython console. Hyperparameter Tuning : The following hyperparameters were configured for the respective machine learning models: Random Forest (RF): Number of estimators = 3, max depth = None. Support Vector Machine (SVM): Kernel = 'RBF', C = 1.0. K-Nearest Neighbors (KNN): Number of neighbors = 3. DBSCAN: Epsilon (eps) = 2, minimum samples = 4. Results Analysis : This section presents a comprehensive analysis and discussion of the performance metrics obtained during evaluation, offering detailed insights into each model's effectiveness in credit card fraud detection. Before delving into the discussion, an overview of the performance parameters used in this study is provided. All metrics, as mentioned in section 4, were derived from the True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) values for each model, as represented in the confusion matrix (CM). The model’s performance was analyzed across the three original and augmented Imbalanced datasets. The results are summarized as follows: Results of Case Study 1 Experiments : Result of first dataset : The 1st imbalanced Dataset (40400) has a test sample 20% (8000 non fraudulent transactions and 80 fraudulent transactions). The test results show that all supervised models (RF, KNN, SVM) achieved zero false positives, demonstrating perfect precision (100%) in correctly identifying all non-fraudulent transactions. Among them, RF was the most effective in detecting fraudulent transactions but misclassified 18 out of 80, resulting in a recall of 78% and an F1 score of 87%. The ensemble majority voting approach, which combines RF, KNN, and SVM, outperformed the individual models, achieving a recall of 79% and an F1 score of 88%, tables (2,3). Figure 3 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Table 2 Performance Comparison Across 3-Models For 1st Original Imbalanced Dataset (40400) Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 0.98 0.78 0.87 KNN 1.00 1.00 0.68 0.81 SVM 1.00 1.00 0.75 0.86 Ensemble voting (RF, KNN&SVM) 1.00 1.00 0.79 0.88 Result of Second Dataset The comparison of results from 2nd imbalanced test samples in Tables 4 , 5 show that all supervised models (RF, KNN, SVM) continue to achieve zero false positives, maintaining perfect precision (100%) in correctly identifying all 16,000 non-fraudulent transactions. Among them, RF remains the top performer in detecting fraudulent transactions, misclassifying 15 out of 160, leading to a recall of 91% and an F1 score of 95%. The ensemble majority voting approach, combining RF, KNN, and SVM, continues to outperform individual models, achieving 100% precision, 92% recall, and an F1 score of 96%. Figure 4 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Table 3 Performance Comparison For 3-Models For 2nd Original imbalanced Dataset (80800) Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 1.00 0.91 0.95 KNN 1.00 1.00 0.81 0.90 SVM 1.00 1.00 0.90 0.95 Ensemble voting (RF, KNN& SVM) 1.00 1.00 0.92 0.96 Result of Third subset The results from the third imbalanced test samples align with the confusion matrix and accuracy values, Tables 6 , 7 . All supervised models (RF, KNN, SVM) maintain a perfect precision of 100% by correctly identifying all 24001 non-fraudulent transactions without any false positives. Among them, RF continues performing the best in detecting fraudulent transactions, misclassifying 19 out of 240, resulting in a recall of 92% and an F1 score of 96%. SVM follows, misclassifying 22 out of 240, with a recall of 91% and an F1 score of 95%. The ensemble majority voting method, combining RF, KNN, and SVM, continues surpassing individual models, achieving 100% precision, 93% recall, and an F1 score of 97%. Figure 5 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Table 4 Performance Comparison For 3-Models For 3rd Original Imbalanced Dataset (121200) Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 0.99 0.92 0.96 KNN 1.00 1.00 0.87 0.93 SVM 1.00 1.00 0.91 0.95 Ensemble voting (RF, KNN&SVM) 1.00 1.00 0.93 0.97 Table 5 Ensemble Majority Voting Performance Across 3- Original Imbalanced Datasets Model Acc. Prec(0) Rec(0) F1-Score(0) Majority vote result (1st dataset) 1.00 1.00 0.79 0.88 Majority vote result (2 nd dataset) 1.00 1.00 0.92 0.96 Majority vote result (3 rd dataset) 1.00 1.00 0.93 0.97 Results Analysis This section presents a comprehensive analysis and discussion of the performance metrics obtained during evaluation, offering detailed insights into each model's effectiveness in credit card fraud detection. Before delving into the discussion, an overview of the performance parameters used in this study is provided. All metrics, as mentioned in section 4, were derived from the True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) values for each model, as represented in the confusion matrix (CM). The model’s performance was analyzed across the three original and augmented Imbalanced datasets. The results are summarized as follows: Results of Case Study 1 Experiments: Result of first dataset: The 1 st imbalanced Dataset (40400) has a test sample 20% (8000 non fraudulent transactions and 80 fraudulent transactions). The test results show that all supervised models (RF, KNN, SVM) achieved zero false positives, demonstrating perfect precision (100%) in correctly identifying all non-fraudulent transactions. Among them, RF was the most effective in detecting fraudulent transactions but misclassified 18 out of 80, resulting in a recall of 78% and an F1 score of 87%. The ensemble majority voting approach, which combines RF, KNN, and SVM, outperformed the individual models, achieving a recall of 79% and an F1 score of 88%, tables (2,3). Figure 3 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 0.98 0.78 0.87 KNN 1.00 1.00 0.68 0.81 SVM 1.00 1.00 0.75 0.86 Ensemble voting (RF, KNN&SVM) 1.00 1.00 0.79 0.88 Table 2. Performance Comparison Across 3-Models For 1st Original Imbalanced Dataset (40400) Result of Second Dataset The comparison of results from 2 nd imbalanced test samples in Tables 4, 5 show that all supervised models (RF, KNN, SVM) continue to achieve zero false positives, maintaining perfect precision (100%) in correctly identifying all 16,000 non-fraudulent transactions. Among them, RF remains the top performer in detecting fraudulent transactions, misclassifying 15 out of 160, leading to a recall of 91% and an F1 score of 95%. The ensemble majority voting approach, combining RF, KNN, and SVM, continues to outperform individual models, achieving 100% precision, 92% recall, and an F1 score of 96%. Figure 4 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 1.00 0.91 0.95 KNN 1.00 1.00 0.81 0.90 SVM 1.00 1.00 0.90 0.95 Ensemble voting (RF, KNN& SVM) 1.00 1.00 0.92 0.96 Table 3. Performance Comparison For 3-Models For 2nd Original imbalanced Dataset (80800) Result of Third subset The results from the third imbalanced test samples align with the confusion matrix and accuracy values, tables 6, 7. All supervised models (RF, KNN, SVM) maintain a perfect precision of 100% by correctly identifying all 24001 non-fraudulent transactions without any false positives. Among them, RF continues performing the best in detecting fraudulent transactions, misclassifying 19 out of 240, resulting in a recall of 92% and an F1 score of 96%. SVM follows, misclassifying 22 out of 240, with a recall of 91% and an F1 score of 95%. The ensemble majority voting method, combining RF, KNN, and SVM, continues surpassing individual models, achieving 100% precision, 93% recall, and an F1 score of 97%. Figure 5 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Model Acc. Prec (0) Rec (0) F1-Score (0) Random Forest 1.00 0.99 0.92 0.96 KNN 1.00 1.00 0.87 0.93 SVM 1.00 1.00 0.91 0.95 Ensemble voting (RF, KNN&SVM) 1.00 1.00 0.93 0.97 Table 4. Performance Comparison For 3-Models For 3rd Original Imbalanced Dataset (121200) Model Acc. Prec(0) Rec(0) F1-Score(0) Majority vote result (1 st dataset) 1.00 1.00 0.79 0.88 Majority vote result (2 nd dataset) 1.00 1.00 0.92 0.96 Majority vote result (3 rd dataset) 1.00 1.00 0.93 0.97 Table 5. Ensemble Majority Voting Performance Across 3- Original Imbalanced Datasets Discussion on the Results of Case Study 1 Ensemble majority voting The results of the testing samples across the three imbalanced datasets in which the ensemble model utilizing majority voting among the three classifiers (SVM, KNN, and RF) outperformed each individual classifier in predicting fraudulent transactions. This approach achieved a recall of up to 93% and an F1 score of up to 97% across all imbalanced datasets, tables 8, 9. Figures 3, 4,5 provide a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart. Discussion on the Results of Case Study 1 Experiments: The results indicate that nearly all supervised models (RF, KNN, SVM) achieve zero false positives across the three imbalanced datasets, demonstrating perfect precision (precision = 1) for the non-fraudulent class. As the dataset size increases, the performance of the supervised models (RF, KNN, SVM) improves in detecting fraudulent transactions, with RF achieving the highest recall of 0.92 across the three datasets. Notably, all three models maintain their precision for the non-fraudulent class (precision = 1). Notably, SVM demonstrates exceptional performance in detecting fraudulent transactions with the largest dataset, achieving a recall of 0.91. This improvement is due to the larger dataset enabling SVM to establish a more generalized and well-defined decision boundary, thereby reducing the risk of misclassifying fraud cases. Although the models achieve perfect accuracy (1.00) across all datasets, their recall is lower than their precision. This suggests that the models fail to identify some fraudulent cases due to class imbalance, where non-fraudulent cases are more prevalent. Consequently, despite maintaining perfect accuracy, the models still misclassify certain fraudulent instances. In the ensemble model, predictions from individual classifiers (RF, KNN, SVM) are combined, with the final prediction determined by the majority vote. This method harnesses the strengths of each model while mitigating their weaknesses, leading to improved accuracy. It demonstrates superiority in detecting fraudulent cases, achieving a recall of up to 93% and an F1 score of up to 97 % across all imbalanced datasets. Results of Case Study 2 Experiments The testing results on the three DBSCAN-augmented imbalanced datasets, utilizing a disjunctive voting strategy among SVM, KNN, and RF classifiers, indicate that the proposed hybrid ensemble model, consistently outperformed the standard ensemble of the same classifiers applied to the original imbalanced datasets without this strategy in detecting fraudulent transactions. This improved approach achieved recall and F1 scores as high as 99.5% and 99.8% respectively, across all datasets. Meanwhile, preserving the perfect accuracy and precision (100%), as shown in Tables 10 and 11. Figure 6 illustrates these outcomes using a bar chart, where each metric (Recall and F1 Score) is represented by a distinct color, with the corresponding legend provided in the top-right corner of the chart. Model Acc. Prec(0) Rec(0) F1-Score(0) Disjunctive vote result (1 st dataset) 1.00 1.00 0.988 0.994 Disjunctive vote result (2 nd dataset) 1.00 1.00 0.994 0.997 Disjunctive vote result (3 rd dataset) 1.00 1.00 0.995 0.998 Table 6. The Proposed Hybrid Ensemble Model Performance Across 3- Datasets Finally, the performance of the proposed hybrid ensemble model is compared with other hybrid models reported in the literature 21, 22 . Table 12 presents the performance metrics using the imbalanced and balanced versions of the European dataset (2013), available on Kaggle. It is observed that balancing the dataset generally enhances the performance of other models. However, improvements in recall often come at the expense of precision. In contrast, the proposed hybrid ensemble model not only outperforms the others but also achieves outstanding results in fraud and non-fraud detection, recording 100% accuracy along with a precision of 100%, recall and F1 scores as high as 99.5% and 99.8% respectively, across all datasets. Reference/year Hybrid Models Accuracy% Precision% Recall% F1-score % Dataset 21 /2022 logistic regression+ Xgoost +  multilayer perceptron 99.98 99.73 94.16 96.86 Imbalanced 100 95.63  99.99 97.76 Balanced 22 / 2022 Random Forest + Adaboost 94 78 85 imbalanced 100 94 97 Balanced The proposed Model RF + SVM + KNN 100 100 99.5 99.8 Imbalanced (augmented with DBSCAN) Table 7 . Comparison of the Proposed Hybrid Ensemble Model with other Hybrid Models in the Literature Conclusions and Future Work A variety of machine learning classifiers, including K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest (RF), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), were assessed as powerful tools for fraud detection. In this study, we performed a comprehensive evaluation of these models using an imbalanced real-world dataset of European credit card transactions. Based on the findings, we developed an ensemble model that combines SVM, KNN, and RF classifiers within a majority voting framework. The results were promising, with the ensemble model demonstrating superior performance compared to individual classifiers—achieving recall rates of up to 93% and F1-scores as high as 97% across all imbalanced datasets. Finally, the paper introduced a hybrid ensemble model that enhances the original imbalanced dataset by applying the unsupervised DBSCAN algorithm to generate cluster labels. These labels are then added as new features, enriching the dataset and providing a more structured data representation. The augmented dataset is subsequently used for training and testing in classification tasks. The hybrid ensemble combines SVM, KNN, and Random Forest (RF) classifiers in a disjunctive voting mechanism specifically designed to better detect fraudulent (outlier) transactions. The results were highly encouraging, as the hybrid ensemble model outperformed all previously evaluated models including other hybrid models reported in the literature — The results were especially compelling, achieving recall and F1-scores as high as 99.5% and 99.8% respectively, and consistently maintaining a perfect accuracy and precision of 100% across all imbalanced datasets. The results confirmed that our hybrid ensemble model successfully minimized false positives and false negatives, addressing two major challenges in credit card fraud detection. However, this study also paves the way for future research. Future research offers numerous avenues for enhancing the model’s efficiency and adaptability. One key area is optimizing the trade-off between accuracy and computational efficiency during both training and testing phases, which could significantly improve overall performance. Minimizing computational overhead would facilitate the deployment of real-time fraud detection systems capable of swiftly responding to emerging fraud patterns. Scalability is another critical aspect—future work could investigate the model’s ability to handle larger datasets and increased computational demands. Techniques such as parallel processing and distributed computing may prove valuable in maintaining performance as data volume grows. Integrating deep learning methods, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), with traditional machine learning approaches could further boost the system's accuracy and flexibility. Additionally, exploring dynamic data sampling strategies that adapt to changes in data distribution over time is vital, particularly in the context of credit card fraud detection, where fraud tactics continually evolve. An adaptive model is more likely to sustain its effectiveness over time. Finally, future studies should also consider enhancing the model’s robustness against novel or adversarial attacks, ensuring greater reliability and security in real-world applications. Declarations Competing interests The authors declare no competing interests. Funding There is no funding for this article Author Contribution Mahmoud A. Ghalwash conducted the experiments and wrote the main manuscript , Samir Mohamed Abdelrazek, Nabila Hamid Aleadawi, and Haitham A. Ghalwash reviewed the manuscript and analyzed the results. Data Availability The Dataset used in this article can be found in Kaggle, Dataset name “Credit Card Fraud Detection Dataset 2023” https://www.kaggle.com/datasets/nelgiriyewithana/credit-card-fraud-detection-dataset-2023?resource=download References Sonam Gupta, Ajay Kumar “A Hybrid Machine Learning Approach for Credit Card Fraud detection”, International Journal of Information Technology Project Management Volume 13, Issue 3, pp. 1-12, 2022. Federal Trade Commission. CSN-Data-Book-2022. no. February 2023. Available online: https://www.ftc.gov/system/files/ftc_gov/pdf/CSN-Data-Book-2022.pdf (accessed on 30 April 2025). UK Finance. Annual Report and Financial Statements 2022. Available online: https://www.ukfinance.org.uk/annual-reports (accessed on 30 April 2025). Gupta, P.; Varshney, A.; Khan, M.R.; Ahmed, R.; Shuaib, M.; Alam, S. “Unbalanced Credit Card Fraud Detection Data: A Machine Learning-Oriented Comparative Study of Balancing Techniques”, Procedia Comput. Sci. 2023 , 218 , 2575–2584. [Elsevier]. Mondal, I.A.; Haque, M.E.; Hassan, A.-M.; Shatabda, S. “Handling imbalanced data for credit card fraud detection”, In Proceedings of the 2021 24th International Conference on Computer and Information Technology (ICCIT), Dhaka, Bangladesh, 18–20 December 2021; pp. 1–6. [Publisher IEEE] Ahmad, H.; Kasasbeh, B.; Aldabaybah, B.; Rawashdeh, E. “Class balancing framework for credit card fraud detection based on clustering and similarity-based selection (SBS)”, Int. J. Inf. Technol. 2023 , ( 15) , 325–333. [ PubMed, https://pubmed.ncbi.nlm.nih.gov] Bagga, S.; Goyal, A.; Gupta, N.; Goyal, A. Credit card fraud detection using pipelining and ensemble learning. Procedia Comput. Sci. 2020 , ( 173) , 104–112. Somvanshi, M.; Chavan, P.; Tambade, S.; Shinde, S.V. A review of machine learning techniques using decision tree and support vector machine. In Proceedings of the 2016 International Conference on Computing Communication Control and Automation (ICCUBEA), Pune, India, 12–13 August 2016; pp. 1–7. (accessed on 30 April 2025). Shah, R. Introduction to k-Nearest Neighbors (kNN) Algorithm. Available online: https://ai.plainenglish.io/introduction-to-k-nearest-neighbors-knn-algorithm-e8617a448fa8 (accessed on 30 April 2025). Randhawa, K.; Loo, C.K.; Seera, M.; Lim, C.P.; Nandi, A.K. Credit card fraud detection using AdaBoost and majority voting. IEEE Access 2018 , 6 , 14277–14284. (accessed on 30 April 2025) Tanouz, D.; Subramanian, R.R.; Eswar, D.; Reddy, G.V.P.; Kumar, A.R.; Praneeth, C.H.V.N.M. Credit Card Fraud Detection Using Machine Learning. In Proceedings of the 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 6–8 May 2021; pp. 967–972. Raghavan, P.; El Gayar, N. Fraud Detection using Machine Learning and Deep Learning. In Proceedings of the 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE), Dubai, United Arab Emirates, 11–12 December 2019; pp. 334–339. Qaddoura, R.; Biltawi, M.M. Improving Fraud Detection in An Imbalanced Class Distribution Using Different Oversampling Techniques. In Proceedings of the 2022 International Engineering Conference on Electrical, Energy, and Artificial Intelligence (EICEEAI), Zarqa, Jordan, 29 November–1 December 2022; pp. 1–5. Sailusha, R.; Gnaneswar, V.; Ramesh, R.; Rao, G.R. Credit Card Fraud Detection Using Machine Learning. In Proceedings of the 2020 4th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 13–15 May 2020; pp. 1264–1270. Jain, Y.; Tiwari, N.; Dubey, S.; Jain, S. A comparative analysis of various credit card fraud detection techniques. Int. J. Recent Technol. Eng. 2019 , 7 , 402–407. Prasad, P.Y.; Chowdary, A.S.; Bavitha, C.; Mounisha, E.; Reethika, C. A Comparison Study of Fraud Detection in Usage of Credit Cards using Machine Learning. In Proceedings of the 2023 7th International Conference on Trends in Electronics and Informatics (ICOEI), Tirunelveli, India, 11–13 April 2023; pp. 1204–1209. Sadgali, I.; Sael, N.; Benabbou, F. Performance of machine learning techniques in the detection of financial frauds. Procedia Comput. Sci. 2019 , 148 , 45–54. Saputra, A.; Suharjito. Fraud detection using machine learning in e-commerce. Int. J. Adv. Comput. Sci. Appl. 2019 , 10 , 332–339. Forough, J.; Momtazi, S. Ensemble of deep sequential models for credit card fraud detection. Appl. Soft Comput. 2021 , ( 99) , 106883. Karthik, V.S.S.; Mishra, A.; Reddy, U.S. Credit card fraud detection by modelling behaviour pattern using hybrid ensemble model. Arab. J. Sci. Eng. 2022 , 47(1) , 1987–1997. Sonam Gupta, A Hybrid Machine Learning Approach for Credit Card Fraud detection, India International Journal of Information Technology Project Management, Volume 13 Issue 3, pp. 1-13, November 2022. Sayali Saraf1, Anupama Phakatkar, Detection of Credit Card Fraud using a Hybrid Ensemble Model, International Journal of Advanced Computer Science and Applications, Vol. 13, No. 9, pp 464-474, 2022. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 13 Nov, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 02 Sep, 2025 Reviews received at journal 23 Aug, 2025 Reviewers agreed at journal 23 Aug, 2025 Reviews received at journal 22 Aug, 2025 Reviewers agreed at journal 22 Aug, 2025 Reviewers invited by journal 17 Aug, 2025 Editor invited by journal 05 Aug, 2025 Editor assigned by journal 01 Aug, 2025 Submission checks completed at journal 31 Jul, 2025 First submitted to journal 28 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7237183","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":504660120,"identity":"22cd9b69-91bd-4750-9a9a-cfea793a6d96","order_by":0,"name":"Mahmoud A. Ghalwash","email":"data:image/png;base64,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","orcid":"","institution":"Mansoura University","correspondingAuthor":true,"prefix":"","firstName":"Mahmoud","middleName":"A.","lastName":"Ghalwash","suffix":""},{"id":504660121,"identity":"93b65ceb-bad4-4ebd-aa39-50f189c5d91a","order_by":1,"name":"Samir Mohamed Abdelrazek","email":"","orcid":"","institution":"Mansoura University","correspondingAuthor":false,"prefix":"","firstName":"Samir","middleName":"Mohamed","lastName":"Abdelrazek","suffix":""},{"id":504660122,"identity":"a0c6c58d-e479-4508-85cc-d46117d15dfb","order_by":2,"name":"Nabila Hamid Eladawi","email":"","orcid":"","institution":"Mansoura University","correspondingAuthor":false,"prefix":"","firstName":"Nabila","middleName":"Hamid","lastName":"Eladawi","suffix":""},{"id":504660123,"identity":"1f21370c-b41e-4d21-a8a6-1e1814beb2a2","order_by":3,"name":"Haitham A. Ghalwash","email":"","orcid":"","institution":"school of Computing Coventry University – Egypt Branch New Cairo","correspondingAuthor":false,"prefix":"","firstName":"Haitham","middleName":"A.","lastName":"Ghalwash","suffix":""}],"badges":[],"createdAt":"2025-07-28 20:53:21","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7237183/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7237183/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-22960-w","type":"published","date":"2025-11-13T15:58:22+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89886036,"identity":"67bcd0ff-6630-4305-b2cf-190954228ead","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":55824,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of the Hybrid Ensemble Model\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/cfbe5b018b396f7573bf23c9.jpg"},{"id":89886035,"identity":"fc36fe96-39b3-41aa-a9f1-421bf6fc927f","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":19429,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of the performance Metrics\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/a8b7586d0168fe9108b9a01b.jpg"},{"id":89886033,"identity":"aabbfcfe-9764-4840-ae8d-105eb3464095","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":44731,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eshows the confusion matrix values for all models based on the testing sample of the First Dataset.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/6a838355a7e8532df846a2ea.jpg"},{"id":89886038,"identity":"50363a82-bdab-4b79-a62f-2b86fd5d1271","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":54040,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of F1-Score \u0026amp; Recall Comparison for the fraud Records for Test Samples of the First Dataset\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/bdea741a87d21d367be4fa2a.jpg"},{"id":89886037,"identity":"937cb0e4-7e5f-4cf7-a0f8-46f3f6f3aa0b","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":48031,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eThe Confusion Matrix Values for All Models Based on the Testing Sample of the Second Dataset.\u003c/em\u003e\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/f150aa7b9d255a0e593b0dcb.jpg"},{"id":89886047,"identity":"68f40ae2-8e16-45ac-9aab-e1aded0867b7","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":63568,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of F1-Score \u0026amp; Recall Comparison for the fraud Records for Test Samples of the Second Dataset\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/5838a8c90b03f4f1985055af.jpg"},{"id":89886051,"identity":"f613aa20-baa1-4781-9764-661f8eb49a50","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":42306,"visible":true,"origin":"","legend":"\u003cp\u003eshows the confusion matrix values for all models based on the testing sample of the third Dataset.\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/67f648353eb6bcf5315b6551.jpg"},{"id":89887202,"identity":"81ae5469-aff6-4db8-a3f1-7ed466e064e1","added_by":"auto","created_at":"2025-08-26 06:52:18","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":67003,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of F1-Score \u0026amp; Recall Comparison for the fraud Records for Test Samples of the Third Dataset.\u003c/p\u003e","description":"","filename":"8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/65882639138846dc50a64a5c.jpg"},{"id":89887203,"identity":"a6a6f6e4-2c73-4905-958f-f76410003e7e","added_by":"auto","created_at":"2025-08-26 06:52:18","extension":"jpg","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":45734,"visible":true,"origin":"","legend":"\u003cp\u003eshows the confusion matrix values for Ensemble Majority voting based on the testing sample Across 3- Original Imbalanced Datasets\u003c/p\u003e","description":"","filename":"9.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/2f2b8723a52184d80880a86d.jpg"},{"id":89886044,"identity":"9dc81198-695d-461b-875b-7cf1e7c1d8cf","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":37962,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eConfusion Matrix for the Proposed Hybrid Ensemble model\u003c/em\u003e\u003c/p\u003e","description":"","filename":"10.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/30c5999ca12b601e85795c37.jpg"},{"id":89886050,"identity":"0dc0a805-6dba-42ec-94e2-2e072dbb7f7c","added_by":"auto","created_at":"2025-08-26 06:36:18","extension":"jpg","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":112361,"visible":true,"origin":"","legend":"\u003cp\u003eIllustration of Recall \u0026amp;F1-Score Comparison between the Ensemble majority Voting \u0026amp; the Proposed Hybrid Ensemble model\u003c/p\u003e","description":"","filename":"11.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/a58846d9f2b417f44c20eb50.jpg"},{"id":96105099,"identity":"671603a0-c8cd-4e78-bbab-79377355c9fc","added_by":"auto","created_at":"2025-11-17 16:08:39","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1829007,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7237183/v1/94c9e5cf-1b38-45cb-93f6-695948f8f146.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Enhancing Credit Card Fraud Detection Using DBSCAN-Augmented Disjunctive Voting Ensemble","fulltext":[{"header":"Introduction","content":"\u003cp\u003eFraudulent activities in the financial sector continue to rise. The use of credit and debit cards for online shopping has significantly increased due to the growth and optimistic outlook of e-commerce. However, this has also led to a heightened risk of credit and debit card fraud \u003csup\u003e1\u003c/sup\u003e. The Federal Trade Commission (FTC) reports that 2021 was a historic year for identity theft, highlighting that fraud losses increase more than 70 percent over 2020 to more than $5.8 billion \u003csup\u003e2\u003c/sup\u003e. The FTC emphasizes the urgent need for innovative solutions to protect both consumers and businesses from these threats. According to the \u003cem\u003eUnited Kingdom Finance Annual Fraud Report 2022\u003c/em\u003e, over £1.3 billion was stolen in 2021 through authorized and unauthorized criminal activities. Despite these challenges, the banking and finance sector prevented an additional £1.4 billion in unauthorized fraud, underscoring the effectiveness of existing measures \u003csup\u003e3\u003c/sup\u003e. In the United States, fraud cases have also surged. The FTC’s \u003cem\u003eConsumer Sentinel Network 2022 Report\u003c/em\u003e recorded 2.4 million fraud complaints in 2022, with total losses reaching nearly $8.8 billion. Investment scams saw the most dramatic rise, with losses reaching nearly $3.8 billion in 2022, more than double the amount reported in 2021 \u003csup\u003e2\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFraud Detection and Ensemble Learning Techniques\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo address these increasing threats, a range of techniques is employed for credit card fraud detection, including statistical, machine learning, and deep learning approaches. Statistical methods such as regression, hypothesis testing, and clustering help identify anomalies in transaction patterns. Machine learning algorithms analyze historical data to detect fraud in real-time, while deep learning utilizes neural networks to uncover intricate patterns in large datasets, delivering high accuracy in fraud detection. A major challenge in credit card fraud detection is the imbalance in data, caused by the uneven distribution of fraudulent and non-fraudulent transactions. This imbalance can result in biased models and diminished effectiveness in identifying fraud. Research \u003csup\u003e4,5,6\u003c/sup\u003e has tackled this issue through approaches such as data balancing, oversampling, under-sampling, and the Synthetic Minority Oversampling Technique (SMOTE). However, a comprehensive evaluation of the effectiveness of these methods is still needed. Ensemble learning techniques, which integrate multiple models, play a crucial role in credit card fraud detection. Approaches such as bagging, boosting, and stacking are especially effective in managing data imbalance and improving predictive performance \u003csup\u003e7\u003c/sup\u003e. By capitalizing on the strengths of various base models, ensemble learning enhances accuracy while minimizing false positives and false negatives.\u003c/p\u003e\n\u003cp\u003eThis paper explores the challenges of credit card fraud detection and provides a review of the state-of-the-art techniques and evaluation criteria. The study aims to propose a framework for a hybrid ensemble of diverse machine learning models, benchmarking its performance against hybrid supervised/unsupervised models.\u003c/p\u003e\n\u003cp\u003eThe primary objective of a fraud detection model is to generate accurate alerts while minimizing false alarms and missed fraud cases. To achieve this, the study conducts a detailed comparative analysis between hybrid supervised/unsupervised models and ensemble models, utilizing various practical evaluation metrics to identify the superior approach for improving credit card fraud detection on transaction data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eObjectives and Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis paper assesses the performance of a hybrid ensemble model that combines multiple algorithms while utilizing imbalanced datasets for credit card fraud detection. The key contributions include:\u003c/p\u003e\n\u003cul\u003e\n\u003cli\u003eAddressing Data Imbalance: Constructing a model tailored to mitigate the issue of disproportionate representation between fraudulent and legitimate transactions.\u003c/li\u003e\n\u003cli\u003eImproving Computational Efficiency: Creating a hybrid ensemble framework optimized for handling intricate algorithms, sophisticated feature engineering, and a variety of base classifiers efficiently.\u003c/li\u003e\n\u003cli\u003eIntroducing a hybrid ensemble approach that leverages Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to refine data features. The ensemble integrates Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) classifiers. Final predictions are determined through a Disjunctive Voting Ensemble (DVE) mechanism. \u003c/li\u003e\n\u003cli\u003ePerformance Evaluation: Comparing the effectiveness of the proposed hybrid ensemble model against individual machine learning algorithms—Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest (RF)—as well as a traditional ensemble model that employs a voting strategy among these classifiers.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe structure of this paper is as follows: Section 2 provides a review of related works, emphasizing machine learning and ensemble techniques for credit card fraud detection. Section 3 presents a detailed explanation of the proposed hybrid ensemble model. Section 4 states the metrics used to evaluate the performance of the models. The experimental setup is comprehensively described in Section 5. Section 6 provides a comprehensive analysis of the experimental results, including performance assessments and comparative evaluations of the proposed hybrid ensemble model. Finally, Section 7 concludes the paper with key findings and insights, along with discussions on future directions for enhancing fraud detection systems.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRelated Work\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis section explores existing literature on credit card fraud detection, with a focus on proposed systems and techniques, particularly Machine Learning and Ensemble Learning models.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine Learning (ML) in Credit Card Fraud Detection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMachine learning algorithms are essential for detecting credit card fraud, as they can analyze data, recognize intricate patterns, and predict fraudulent transactions. These algorithms fall into two main categories: supervised and unsupervised learning methods. Commonly used techniques for Credit Card Fraud Detection (CCFD) include Logistic Regression (LR), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Naive Bayes (NB), Decision Trees (DT), Random Forest (RF), and others. Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Random Forest (RF), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN) classifiers (supervised and unsupervised), are among the most powerful machine learning models (used in fraud detection. SVM classifies data by finding the optimal hyperplane \u003csup\u003e8\u003c/sup\u003e, KNN classifies transactions based on the nearest neighbors \u003csup\u003e9\u003c/sup\u003e, RF aggregates decision trees to reduce overfitting \u003csup\u003e10\u003c/sup\u003e, and DBSCAN. These diverse approaches contribute to the robustness of fraud detection systems, offering effective ways to identify and prevent fraudulent transactions. \u003c/p\u003e\n\u003cp\u003eTanouz et al.\u003csup\u003e11\u003c/sup\u003e carried out an in-depth study on machine learning techniques for credit card fraud classification, with a particular emphasis on imbalanced datasets. Their findings demonstrated that Random Forest is a robust approach for fraud detection. However, the absence of a feature selection process constrained the models' performance.\u003c/p\u003e\n\u003cp\u003eRaghavan et al.\u003csup\u003e12\u003c/sup\u003e investigated fraud detection by applying data mining techniques to three datasets from Australia (AU), Germany, and Europe (EU). Their study utilized algorithms such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Random Forest. Additionally, they developed two ensemble models: one integrating KNN, SVM, and Convolutional Neural Network (CNN), and another combining KNN, SVM, and Random Forest. The findings revealed that SVM outperformed the other algorithms. While their research provided valuable insights into the effectiveness of various algorithms and ensemble approaches for fraud detection, overall performance remained relatively low across all datasets. \u003c/p\u003e\n\u003cp\u003eIn 2022, Qaddoura et al.\u003csup\u003e13\u003c/sup\u003e examined the impact of different oversampling techniques, including SMOTE, ADASYN, borderline1, borderline2, and SVM-based oversampling algorithms. Their study revealed that applying oversampling methods can significantly improve model performance. \u003c/p\u003e\n\u003cp\u003eIn 2022, Sahithi et al. [1] presented a credit card fraud detection model based on a Weighted Average Ensemble, integrating LR, RF, KNN, Adaboost, and Bagging. Their study highlights the effectiveness of ensemble models in detecting credit card fraud within this critical domain. However, the limited discussion on the feature selection process affects the model's reproducibility.\u003c/p\u003e\n\u003cp\u003eRuttala et al. \u003csup\u003e14\u003c/sup\u003e conducted a comparative analysis of the Random Forest and AdaBoost algorithms for credit card fraud detection using an imbalanced dataset. Their findings revealed that Random Forest outperformed AdaBoost in terms of precision, recall, and F1-score.\u003c/p\u003e\n\u003cp\u003eTiwari et al. \u003csup\u003e15\u003c/sup\u003e conducted a comparative study of various credit card fraud detection techniques, assessing algorithms such as SVM, ANN, Bayesian Network, K-Nearest Neighbor (KNN), and Decision Trees. Using the KDD dataset from the KDD CUP 99 Intrusion Dataset, they found varying accuracy levels: SVM achieved 94.65%, ANN reached 99.71%, KNN attained 97.15%, and Decision Trees recorded 94.7%. While their analysis provided valuable insights into fraud detection methods, its effectiveness was limited by the dataset’s inability to fully represent real-world financial activities. Numerous studies have investigated methods to enhance fraud prevention and detection in credit card transactions using machine learning.\u003c/p\u003e\n\u003cp\u003ePrasad Chowdary et al \u003csup\u003e16\u003c/sup\u003e. introduced an ensemble approach to improve CCFD. Their work emphasizes optimizing model parameters, improving performance metrics, and incorporating deep learning to minimize identification errors and false negatives. By combining multiple classifiers and conducting rigorous evaluations, their approach enhances the efficiency of CCFD systems.\u003c/p\u003e\n\u003cp\u003eSadgali et al. \u003csup\u003e17\u003c/sup\u003e aimed to determine the most effective techniques for detecting financial fraud. Their approach incorporated various methods, including Support Vector Machine (SVM). Notably, their study did not focus on a specific dataset for analysis. The results indicated that Naïve Bayes achieved the highest performance, with SVM closely following. However, the research was limited to insurance fraud detection.\u003c/p\u003e\n\u003cp\u003eSaputra et al. \u003csup\u003e18\u003c/sup\u003e evaluated the performance of Decision Tree, Naïve Bayes, Random Forest, and Neural Network algorithms for fraud detection, utilizing SMOTE to mitigate dataset imbalance. The Kaggle dataset used in the study contained a low percentage of fraudulent transactions (0.093%). Based on confusion matrix analysis, the results showed that the Neural Network achieved the highest accuracy, followed by Random Forest. Additionally, SMOTE significantly improved the average F1-score, effectively handling the data imbalance. \u003c/p\u003e\n\u003cp\u003eForough et al. \u003csup\u003e19\u003c/sup\u003e developed an ensemble model that integrates deep recurrent neural networks with an innovative voting mechanism based on an artificial neural network to detect fraudulent activities. The model employs multiple recurrent networks, such as LSTM or GRU, as base classifiers, and combines their outputs using a feed-forward neural network (FFNN) for the voting process. The ensemble model, utilizing GRU, achieves its best performance with two base classifiers on both the European cards dataset and the Brazilian dataset. It outperforms the individual GRU model across all metrics and the baseline ensemble model in most metrics. \u003c/p\u003e\n\u003cp\u003eKarthik et al. \u003csup\u003e20\u003c/sup\u003e proposed a novel credit card fraud detection model that combines ensemble learning techniques such as boosting and bagging. This hybrid classifier leverages the strengths of both methods, with Adaboost used for feature engineering of the behavioral feature space. The model's performance was assessed using the area under the precision-recall (AUPR) curve, showing moderate improvements, with results ranging from 58.03% to 69.97% on the Brazilian bank dataset and from 54.66% to 69.40% on the UCSD-FICO dataset.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe Proposed Hybrid Ensemble Framework\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eThe Proposed Hybrid Ensemble Model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe proposed hybrid ensemble model integrates Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) with feature engineering using DBSCAN. As an unsupervised clustering algorithm, DBSCAN detects anomalies and patterns within the dataset. Its outputs are incorporated as an additional feature, creating an augmented dataset that enhances the information available to the supervised classifiers. By applying the ensemble method to the augmented dataset, the model leverages the distinct advantages of its components: the robustness of Random Forest (RF), the simplicity of K-Nearest Neighbors (KNN), and the ability of Support Vector Machine (SVM) to handle complex decision boundaries. The integration of DBSCAN enhances the feature set by combining supervised learning with unsupervised clustering, which boosts both predictive accuracy and model robustness—particularly valuable in fraud detection scenarios. Furthermore, the use of disjunctive voting emphasizes recall, reducing the likelihood of overlooking fraudulent transactions and resulting in a more dependable system for fraud detection and other classification tasks.\u003c/p\u003e"},{"header":"Methodology","content":"\u003cp\u003e\u003cb\u003eThe Classification Phase\u003c/b\u003e:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eThe proposed\u003c/b\u003e hybrid ensemble model \u003cb\u003eintegrates three key algorithms\u003c/b\u003e:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eدRandom Forest (RF): A supervised ensemble method that constructs multiple decision trees and aggregates their predictions to enhance accuracy.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSupport Vector Machine (SVM): A supervised learning model that identifies the optimal hyperplane to separate fraud and non-fraud transactions.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eK-Nearest Neighbors (KNN): A distance-based algorithm that classifies a sample based on the majority class among its nearest neighbors.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eDisjunctive Voting Mechanism\u003c/b\u003e\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eIn a Disjunctive (OR-based) voting scheme, a transaction is classified as fraudulent if \u003cb\u003eany\u003c/b\u003e of the individual classifiers predicts it as fraud. This is advantageous in imbalanced settings where fraudulent transactions (minority class) are rare but critical to detect. It increases the \u003cb\u003esensitivity\u003c/b\u003e (recall) by ensuring that potential frauds are less likely to be missed. The predictions from RF, SVM, and KNN are combined using \u003cb\u003ea\u003c/b\u003e disjunctive voting approach, where each algorithm's output contributes to the final classification decision.\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eThe Ensemble Disjunctive Voting Mechanism\u003c/b\u003e:\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eThe disjunctive voting mechanism among the three integrated algorithms (\u003cb\u003eRF, KNN, and SVM\u003c/b\u003e) determines the final classification decision, as outlined in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eIf RF, KNN, and SVM unanimously vote \"non-fraud\", the final decision of the ensemble model will be \u003cb\u003e\"\u003c/b\u003enon-fraud\u003cb\u003e\"\u003c/b\u003e.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eOtherwise, if any of the models vote \u0026ldquo;fraud\u0026rdquo; on a classification, this determines the final decision.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003ePerformance Metrics\u003c/b\u003e:\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eEnsemble Disjunctive Voting Decisions\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModels\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e\u003cp\u003eClassification\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNon-fraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNon-fraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNon-fraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003ex\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFinal Classification\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eNon-fraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003efraud\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eThe following metrics are used to evaluate the performance of the models:\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eAccuracy: Measures the overall performance of the model by calculating the percentage of correctly classified instances out of the total examined cases.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003ePrecision: Defined as the ratio of correctly predicted positive instances to the total predicted positives, precision assesses the accuracy of the model\u0026rsquo;s positive predictions. This metric is particularly important in scenarios \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ewhere false positives carry significant consequences.\u003c/span\u003e\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eRecall (Sensitivity): Represents the ratio of correctly predicted positive instances to the total actual positives, indicating the model\u0026rsquo;s ability to identify all relevant occurrences. High recall is crucial in applications \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003ewere failing to detect positive cases could lead to serious consequences.\u003c/span\u003e\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eF1 Score: A harmonic mean of precision and recall, offering a balanced evaluation of both metrics. It is particularly useful in cases with \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eimbalanced class distributions\u003c/span\u003e, where \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003eaccuracy alone may not provide a complete picture of model performance\u003c/span\u003e.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eConfusion Matrix: Provides a breakdown of model predictions into true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN), enabling a detailed performance analysis.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eConfusion Matrix\u003c/b\u003e: (5)\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eFigure 2.\u003c/b\u003e \u003cem\u003eIllustration of the performance Metrics\u003c/em\u003e\u003c/p\u003e\u003cp\u003eIt is important to note that the F1 score is particularly useful when accuracy is misleading, especially in imbalanced datasets. It effectively balances precision and recall, addressing the following trade-offs:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eHigh precision, low recall \u0026rarr; The model is too strict, resulting in fewer false positives but many false negatives.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eHigh recall, low precision \u0026rarr; The model is too lenient, leading to more false positives but fewer false negatives.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eIn credit card fraud detection, the goal is to achieve both high precision and recall to effectively prevent fraudulent transactions while maintaining system performance by accurately accepting legitimate ones.\u003c/p\u003e\u003cp\u003e\u003cb\u003eExperimental Setup\u003c/b\u003e:\u003c/p\u003e\u003cp\u003e\u003cb\u003eCase Study 1 Using The Original Imbalanced Dataset\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe experiments are conducted on the original subset of 242,400 records that is split into three imbalanced subsets (40400, 80800, and 121200 records) as mentioned in section 3.2. The case study includes two Experiments.\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eTrain and test individual algorithms\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eEach individual algorithm\u0026mdash;RF, KNN, and SVM\u0026mdash;is trained and tested on the three imbalanced datasets using an 80\u0026thinsp;\u0026minus;\u0026thinsp;20 split. Their effectiveness in detecting fraud (0) and non-fraud transactions (1) is measured using the performance metrics described in Section 4.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eEnsemble majority evaluation\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003eAdditionally, a majority voting strategy is applied to the three ensemble supervised models (RF, KNN, SVM) trained on the imbalanced datasets. The ensemble model\u0026rsquo;s performance in identifying fraudulent (0) and non-fraudulent (1) transactions is also evaluated using the performance metrics outlined in Section 4.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe performance of individual algorithms (RF, KNN, and SVM) is compared with the majority voting strategy of the three ensemble supervised models (RF, KNN, SVM) to evaluate their effectiveness in detecting both fraudulent and non-fraudulent credit card transactions.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eCase Study 2 Applying The Proposed Hybrid Ensemble Model\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe experiments are carried out on the augmented dataset consisting of 242,400 records, which is divided into three imbalanced subsets (40,400, 80,800, and 121,200 records) as described in Section 3.4. The case study follows a two-step process:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eFirst, individual algorithms\u0026mdash;RF, KNN, and SVM\u0026mdash;are trained and tested on the three augmented imbalanced datasets using an 80\u0026thinsp;\u0026minus;\u0026thinsp;20 split. Their ability to classify fraud and non-fraud transactions is evaluated and incorporated into the disjunctive voting mechanism described in section 3.5.2.1.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSecond, the disjunctive voting mechanism for the proposed hybrid ensemble model, as described in Section 3.5.2.1, is applied to the classifications produced by individual algorithms (RF, KNN, and SVM) for final evaluation. The performance of this hybrid ensemble model in identifying fraudulent and non-fraudulent transactions is measured using the metrics specified in Section 4.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eThe performance results of the proposed Hybrid Ensemble Model are documented and compared with those of other models to draw a conclusion.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eImplementation Setup\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe models were implemented using the Spyder application for Python programming. The development process utilized Python along with libraries such as Scikit-learn, Pandas, and NumPy. Data preprocessing, model training, and evaluation were conducted within the IPython console.\u003c/p\u003e\u003cp\u003e\u003cb\u003eHyperparameter Tuning\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe following hyperparameters were configured for the respective machine learning models:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eRandom Forest (RF): Number of estimators\u0026thinsp;=\u0026thinsp;3, max depth\u0026thinsp;=\u0026thinsp;None.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eSupport Vector Machine (SVM): Kernel = 'RBF', C\u0026thinsp;=\u0026thinsp;1.0.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eK-Nearest Neighbors (KNN): Number of neighbors\u0026thinsp;=\u0026thinsp;3.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eDBSCAN: Epsilon (eps)\u0026thinsp;=\u0026thinsp;2, minimum samples\u0026thinsp;=\u0026thinsp;4.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults Analysis\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThis section presents a comprehensive analysis and discussion of the performance metrics obtained during evaluation, offering detailed insights into each model's effectiveness in credit card fraud detection. Before delving into the discussion, an overview of the performance parameters used in this study is provided. All metrics, as mentioned in section 4, were derived from the True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) values for each model, as represented in the confusion matrix (CM).\u003c/p\u003e\u003cp\u003eThe model\u0026rsquo;s performance was analyzed across the three original and augmented Imbalanced datasets. The results are summarized as follows:\u003c/p\u003e\u003cp\u003e\u003cb\u003eResults of Case Study 1 Experiments\u003c/b\u003e:\u003c/p\u003e\u003cp\u003e\u003cb\u003eResult of first dataset\u003c/b\u003e:\u003c/p\u003e\u003cp\u003eThe 1st imbalanced Dataset (40400) has a test sample 20% (8000 non fraudulent transactions and 80 fraudulent transactions). The test results show that all supervised models (RF, KNN, SVM) achieved zero false positives, demonstrating perfect precision (100%) in correctly identifying all non-fraudulent transactions. Among them, RF was the most effective in detecting fraudulent transactions but misclassified 18 out of 80, resulting in a recall of 78% and an F1 score of 87%. The ensemble majority voting approach, which combines RF, KNN, and SVM, outperformed the individual models, achieving a recall of 79% and an F1 score of 88%, tables (2,3). Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003e provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance Comparison Across 3-Models For 1st Original Imbalanced Dataset (40400)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAcc.\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1-Score (0)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.68\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.81\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.75\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.86\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEnsemble voting (RF, KNN\u0026amp;SVM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eResult of Second Dataset\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe comparison of results from 2nd imbalanced test samples in Tables\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, \u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e show that all supervised models (RF, KNN, SVM) continue to achieve zero false positives, maintaining perfect precision (100%) in correctly identifying all 16,000 non-fraudulent transactions. Among them, RF remains the top performer in detecting fraudulent transactions, misclassifying 15 out of 160, leading to a recall of 91% and an F1 score of 95%. The ensemble majority voting approach, combining RF, KNN, and SVM, continues to outperform individual models, achieving 100% precision, 92% recall, and an F1 score of 96%. Figure\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003e provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance Comparison For 3-Models For 2nd Original imbalanced Dataset (80800)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAcc.\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1-Score (0)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.81\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.90\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.90\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEnsemble voting (RF, KNN\u0026amp; SVM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003eResult of Third subset\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe results from the third imbalanced test samples align with the confusion matrix and accuracy values, Tables \u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e, \u003cspan refid=\"Tab7\" class=\"InternalRef\"\u003e7\u003c/span\u003e. All supervised models (RF, KNN, SVM) maintain a perfect precision of 100% by correctly identifying all 24001 non-fraudulent transactions without any false positives. Among them, RF continues performing the best in detecting fraudulent transactions, misclassifying 19 out of 240, resulting in a recall of 92% and an F1 score of 96%. SVM follows, misclassifying 22 out of 240, with a recall of 91% and an F1 score of 95%. The ensemble majority voting method, combining RF, KNN, and SVM, continues surpassing individual models, achieving 100% precision, 93% recall, and an F1 score of 97%. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePerformance Comparison For 3-Models For 3rd Original Imbalanced Dataset (121200)\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAcc.\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRec (0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1-Score (0)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e0.99\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eKNN\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.87\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.91\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.95\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEnsemble voting (RF, KNN\u0026amp;SVM)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eEnsemble Majority Voting Performance Across 3- Original Imbalanced Datasets\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAcc.\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003ePrec(0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eRec(0)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eF1-Score(0)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMajority vote result (1st dataset)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.79\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.88\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMajority vote result (2\u003csub\u003end\u003c/sub\u003e dataset)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.92\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.96\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMajority vote result (3\u003csub\u003erd\u003c/sub\u003e dataset)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e1.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e0.93\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.97\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Results Analysis","content":"\u003cp\u003eThis section presents a comprehensive analysis and discussion of the performance metrics obtained during evaluation, offering detailed insights into each model\u0026apos;s effectiveness in credit card fraud detection. Before delving into the discussion, an overview of the performance parameters used in this study is provided. All metrics, as mentioned in section 4, were derived from the True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) values for each model, as represented in the confusion matrix (CM).\u003c/p\u003e\n\u003cp\u003eThe model\u0026rsquo;s performance was analyzed across the three original and augmented Imbalanced datasets. The results are summarized as follows:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u0026nbsp;of Case Study 1 Experiments:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;Result of first dataset:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe 1\u003csup\u003est\u003c/sup\u003e imbalanced Dataset (40400) has a test sample 20% (8000 non fraudulent transactions and 80 fraudulent transactions). The test results show that all supervised models (RF, KNN, SVM) achieved zero false positives, demonstrating perfect precision (100%) in correctly identifying all non-fraudulent transactions. Among them, RF was the most effective in detecting fraudulent transactions but misclassified 18 out of 80, resulting in a recall of 78% and an F1 score of 87%. The ensemble majority voting approach, which combines RF, KNN, and SVM, outperformed the individual models, achieving a recall of 79% and an F1 score of 88%, tables (2,3). Figure 3 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" align=\"\" width=\"441\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Rec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF1-Score (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEnsemble voting (RF, KNN\u0026amp;SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;0.79 \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 2. Performance Comparison Across 3-Models For 1st Original Imbalanced Dataset (40400)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResult of Second Dataset\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe comparison of results from 2\u003csup\u003end\u003c/sup\u003e imbalanced test samples in Tables 4, 5 show that all supervised models (RF, KNN, SVM) continue to achieve zero false positives, maintaining perfect precision (100%) in correctly identifying all 16,000 non-fraudulent transactions. Among them, RF remains the top performer in detecting fraudulent transactions, misclassifying 15 out of 160, leading to a recall of 91% and an F1 score of 95%. The ensemble majority voting approach, combining RF, KNN, and SVM, continues to outperform individual models, achieving 100% precision, 92% recall, and an F1 score of 96%. Figure 4 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"437\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Rec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF1-Score (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEnsemble voting (RF, KNN\u0026amp; SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 3. Performance Comparison For 3-Models For 2nd Original imbalanced Dataset (80800)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResult of Third subset\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe results from the third imbalanced test samples align with the confusion matrix and accuracy values, tables 6, 7. All supervised models (RF, KNN, SVM) maintain a perfect precision of 100% by correctly identifying all 24001 non-fraudulent transactions without any false positives. Among them, RF continues performing the best in detecting fraudulent transactions, misclassifying 19 out of 240, resulting in a recall of 92% and an F1 score of 96%. SVM follows, misclassifying 22 out of 240, with a recall of 91% and an F1 score of 95%. The ensemble majority voting method, combining RF, KNN, and SVM, continues surpassing individual models, achieving 100% precision, 93% recall, and an F1 score of 97%. \u0026nbsp;Figure 5 provides a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"419\" class=\"fr-table-selection-hover\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5584%;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Rec (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6612%;\"\u003e\n \u003cp\u003eF1-Score (0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5584%;\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6612%;\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5584%;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6612%;\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5584%;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6612%;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEnsemble voting (RF, KNN\u0026amp;SVM)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14.5584%;\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 17.6612%;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable\u0026nbsp;4.\u003c/strong\u003e Performance Comparison For 3-Models For 3rd Original Imbalanced Dataset (121200)\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"454\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrec(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Rec(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF1-Score(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMajority vote result (1\u003csup\u003est\u003c/sup\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.79\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMajority vote result (2\u003csub\u003end\u003c/sub\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eMajority vote result (3\u003csub\u003erd\u003c/sub\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTable 5. Ensemble Majority Voting Performance Across 3- Original Imbalanced Datasets\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Discussion on the Results of Case Study 1 Ensemble majority voting","content":"\u003cp\u003eThe results of the testing samples across the three imbalanced datasets in which the ensemble model utilizing majority voting among the three classifiers (SVM, KNN, and RF) outperformed each individual classifier in predicting fraudulent transactions. This approach achieved a recall of up to 93% and an F1 score of up to 97% across all imbalanced datasets, tables 8, 9. Figures 3, 4,5 provide a visual representation of these values in a bar chart, where each color represents a specific metric (Recall, F1 Score). The color -to- metric mapping is detailed in the legend box at the top right corner of the chart.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDiscussion on the Results of Case Study 1 Experiments:\u003c/strong\u003e\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eThe results indicate that nearly all supervised models (RF, KNN, SVM) achieve zero false positives across the three imbalanced datasets, demonstrating perfect precision (precision = 1) for the non-fraudulent class.\u003c/li\u003e\n \u003cli\u003eAs the dataset size increases, the performance of the supervised models (RF, KNN, SVM) improves in detecting fraudulent transactions, with RF achieving the highest recall of 0.92 across the three datasets. Notably, all three models maintain their precision for the non-fraudulent class (precision = 1).\u003c/li\u003e\n \u003cli\u003eNotably, SVM demonstrates exceptional performance in detecting fraudulent transactions with the largest dataset, achieving a recall of 0.91. This improvement is due to the larger dataset enabling SVM to establish a more generalized and well-defined decision boundary, thereby reducing the risk of misclassifying fraud cases.\u003c/li\u003e\n \u003cli\u003eAlthough the models achieve perfect accuracy (1.00) across all datasets, their recall is lower than their precision. This suggests that the models fail to identify some fraudulent cases due to class imbalance, where non-fraudulent cases are more prevalent. Consequently, despite maintaining perfect accuracy, the models still misclassify certain fraudulent instances.\u003c/li\u003e\n \u003cli\u003eIn the ensemble model, predictions from individual classifiers (RF, KNN, SVM) are combined, with the final prediction determined by the majority vote. This method harnesses the strengths of each model while mitigating their weaknesses, leading to improved accuracy. It demonstrates superiority in detecting fraudulent cases, achieving a recall of up to 93% and an F1 score of up to 97 % across all imbalanced datasets.\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults of Case Study 2 Experiments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe testing results on the three DBSCAN-augmented imbalanced datasets, utilizing a disjunctive voting strategy among SVM, KNN, and RF classifiers, indicate that the proposed hybrid ensemble model, consistently outperformed the standard ensemble of the same classifiers applied to the original imbalanced datasets without this strategy in detecting fraudulent transactions. This improved approach achieved recall and F1 scores as high as 99.5% and 99.8% respectively, across all datasets. Meanwhile, preserving the perfect accuracy and precision (100%), as shown in Tables 10 and 11. Figure 6 illustrates these outcomes using a bar chart, where each metric (Recall and F1 Score) is represented by a distinct color, with the corresponding legend provided in the top-right corner of the chart.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"454\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eModel\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAcc.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrec(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRec(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF1-Score(0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDisjunctive vote result (1\u003csup\u003est\u003c/sup\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.988\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.994\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDisjunctive vote result (2\u003csub\u003end\u003c/sub\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.994\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.997\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDisjunctive vote result (3\u003csub\u003erd\u003c/sub\u003e dataset)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.995\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.998\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTable 6.\u003c/em\u003e\u003c/strong\u003e\u003cem\u003eThe Proposed Hybrid Ensemble Model Performance Across 3- Datasets\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eFinally, the performance of the proposed hybrid ensemble model is compared with other hybrid models reported in the literature \u003csup\u003e21, 22\u003c/sup\u003e. Table 12 presents the performance metrics using the imbalanced and balanced versions of the European dataset (2013), available on Kaggle. It is observed that balancing the dataset generally enhances the performance of other models. However, improvements in recall often come at the expense of precision. In contrast, the proposed hybrid ensemble model not only outperforms the others but also achieves outstanding results in fraud and non-fraud detection, recording 100% accuracy along with a precision of 100%, recall and F1 scores as high as 99.5% and 99.8% respectively, across all datasets.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"636\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eReference/year\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHybrid Models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAccuracy%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrecision%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRecall%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eF1-score %\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDataset\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u003csup\u003e21\u003c/sup\u003e/2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003elogistic regression+ Xgoost +  multilayer\u003c/p\u003e\n \u003cp\u003eperceptron\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e99.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e99.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e94.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eImbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cu\u003e100\u003c/u\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e95.63\u003c/strong\u003e \u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cu\u003e99.99\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e97.76\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u003csup\u003e22\u003c/sup\u003e/ 2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd rowspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003eRandom Forest + Adaboost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eimbalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBalanced\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eThe proposed Model\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRF + SVM + KNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cu\u003e100\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cu\u003e100\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cu\u003e99.5\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u003cu\u003e99.8\u003c/u\u003e\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cu\u003eImbalanced (augmented with DBSCAN)\u003c/u\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eTable 7\u003c/em\u003e\u003c/strong\u003e\u003cem\u003e. Comparison of the Proposed Hybrid Ensemble Model with other Hybrid Models in the Literature\u003c/em\u003e\u003c/p\u003e"},{"header":"Conclusions and Future Work","content":"\u003cp\u003eA variety of machine learning classifiers, including K-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest (RF), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), were assessed as powerful tools for fraud detection. In this study, we performed a comprehensive evaluation of these models using an imbalanced real-world dataset of European credit card transactions. Based on the findings, we developed an ensemble model that combines SVM, KNN, and RF classifiers within a majority voting framework. The results were promising, with the ensemble model demonstrating superior performance compared to individual classifiers\u0026mdash;achieving recall rates of up to 93% and F1-scores as high as 97% across all imbalanced datasets.\u003c/p\u003e\n\u003cp\u003eFinally, the paper introduced a hybrid ensemble model that enhances the original imbalanced dataset by applying the unsupervised DBSCAN algorithm to generate cluster labels. These labels are then added as new features, enriching the dataset and providing a more structured data representation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe augmented dataset is subsequently used for training and testing in classification tasks. The hybrid ensemble combines SVM, KNN, and Random Forest (RF) classifiers in a disjunctive voting mechanism specifically designed to better detect fraudulent (outlier) transactions. The results were highly encouraging, as the hybrid ensemble model outperformed all previously evaluated models including other hybrid models reported in the literature \u0026mdash; The results were especially compelling, achieving recall and F1-scores as high as 99.5% and 99.8% respectively, and consistently maintaining a perfect accuracy and precision of 100% across all imbalanced datasets.\u003c/p\u003e\n\u003cp\u003eThe results confirmed that our hybrid ensemble model successfully minimized false positives and false negatives, addressing two major challenges in credit card fraud detection. However, this study also paves the way for future research.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFuture research offers numerous avenues for enhancing the model\u0026rsquo;s efficiency and adaptability. One key area is optimizing the trade-off between accuracy and computational efficiency during both training and testing phases, which could significantly improve overall performance. Minimizing computational overhead would facilitate the deployment of real-time fraud detection systems capable of swiftly responding to emerging fraud patterns. Scalability is another critical aspect\u0026mdash;future work could investigate the model\u0026rsquo;s ability to handle larger datasets and increased computational demands. Techniques such as parallel processing and distributed computing may prove valuable in maintaining performance as data volume grows.\u003c/p\u003e\n\u003cp\u003eIntegrating deep learning methods, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), with traditional machine learning approaches could further boost the system\u0026apos;s accuracy and flexibility. Additionally, exploring dynamic data sampling strategies that adapt to changes in data distribution over time is vital, particularly in the context of credit card fraud detection, where fraud tactics continually evolve. An adaptive model is more likely to sustain its effectiveness over time.\u003c/p\u003e\n\u003cp\u003eFinally, future studies should also consider enhancing the model\u0026rsquo;s robustness against novel or adversarial attacks, ensuring greater reliability and security in real-world applications.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThere is no funding for this article\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eMahmoud A. Ghalwash conducted the experiments and wrote the main manuscript , Samir Mohamed Abdelrazek, Nabila Hamid Aleadawi, and Haitham A. Ghalwash reviewed the manuscript and analyzed the results.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe Dataset used in this article can be found in Kaggle, Dataset name \u0026ldquo;Credit Card Fraud Detection Dataset 2023\u0026rdquo; https://www.kaggle.com/datasets/nelgiriyewithana/credit-card-fraud-detection-dataset-2023?resource=download\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eSonam Gupta, Ajay Kumar \u0026ldquo;A Hybrid Machine Learning Approach for Credit Card Fraud detection\u0026rdquo;, International Journal of Information Technology Project Management Volume 13, Issue 3, pp. 1-12, 2022. \u003c/li\u003e\n\u003cli\u003eFederal Trade Commission. CSN-Data-Book-2022. no. February 2023. Available online: \u003cstrong\u003ehttps://www.ftc.gov/system/files/ftc_gov/pdf/CSN-Data-Book-2022.pdf\u003c/strong\u003e (accessed on 30 April 2025).\u003c/li\u003e\n\u003cli\u003eUK Finance. Annual Report and Financial Statements 2022. Available online: \u003cstrong\u003ehttps://www.ukfinance.org.uk/annual-reports\u003c/strong\u003e (accessed on 30 April 2025). \u003c/li\u003e\n\u003cli\u003eGupta, P.; Varshney, A.; Khan, M.R.; Ahmed, R.; Shuaib, M.; Alam, S. \u0026ldquo;Unbalanced Credit Card Fraud Detection Data: A Machine Learning-Oriented Comparative Study of Balancing Techniques\u0026rdquo;, Procedia\u003cem\u003e Comput. Sci.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, \u003cem\u003e218\u003c/em\u003e, 2575\u0026ndash;2584. [Elsevier].\u003c/li\u003e\n\u003cli\u003eMondal, I.A.; Haque, M.E.; Hassan, A.-M.; Shatabda, S. \u0026ldquo;Handling imbalanced data for credit card fraud detection\u0026rdquo;, In Proceedings of the 2021 24th International Conference on Computer and Information Technology (ICCIT), Dhaka, Bangladesh, 18\u0026ndash;20 December 2021; pp. 1\u0026ndash;6. [Publisher IEEE]\u003c/li\u003e\n\u003cli\u003eAhmad, H.; Kasasbeh, B.; Aldabaybah, B.; Rawashdeh, E. \u0026ldquo;Class balancing framework for credit card fraud detection based on clustering and similarity-based selection (SBS)\u0026rdquo;, \u003cem\u003eInt. J. Inf. Technol.\u003c/em\u003e\u003cstrong\u003e2023\u003c/strong\u003e, (\u003cem\u003e15)\u003c/em\u003e, 325\u0026ndash;333. [ PubMed, https://pubmed.ncbi.nlm.nih.gov] \u003c/li\u003e\n\u003cli\u003eBagga, S.; Goyal, A.; Gupta, N.; Goyal, A. Credit card fraud detection using pipelining and ensemble learning. \u003cem\u003eProcedia Comput. Sci.\u003c/em\u003e\u003cstrong\u003e2020\u003c/strong\u003e, (\u003cem\u003e173)\u003c/em\u003e, 104\u0026ndash;112. \u003c/li\u003e\n\u003cli\u003eSomvanshi, M.; Chavan, P.; Tambade, S.; Shinde, S.V. A review of machine learning techniques using decision tree and support vector machine. In Proceedings of the 2016 International Conference on Computing Communication Control and Automation (ICCUBEA), Pune, India, 12\u0026ndash;13 August 2016; pp. 1\u0026ndash;7. (accessed on 30 April 2025).\u003c/li\u003e\n\u003cli\u003eShah, R. Introduction to k-Nearest Neighbors (kNN) Algorithm. Available online: \u003cstrong\u003ehttps://ai.plainenglish.io/introduction-to-k-nearest-neighbors-knn-algorithm-e8617a448fa8\u003c/strong\u003e (accessed on 30 April 2025).\u003c/li\u003e\n\u003cli\u003eRandhawa, K.; Loo, C.K.; Seera, M.; Lim, C.P.; Nandi, A.K. Credit card fraud detection using AdaBoost and majority voting. \u003cem\u003eIEEE Access\u003c/em\u003e\u003cstrong\u003e2018\u003c/strong\u003e, \u003cem\u003e6\u003c/em\u003e, 14277\u0026ndash;14284. (accessed on 30 April 2025)\u003c/li\u003e\n\u003cli\u003eTanouz, D.; Subramanian, R.R.; Eswar, D.; Reddy, G.V.P.; Kumar, A.R.; Praneeth, C.H.V.N.M. Credit Card Fraud Detection Using Machine Learning. In Proceedings of the 2021 5th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 6\u0026ndash;8 May 2021; pp. 967\u0026ndash;972. \u003c/li\u003e\n\u003cli\u003eRaghavan, P.; El Gayar, N. Fraud Detection using Machine Learning and Deep Learning. In Proceedings of the 2019 International Conference on Computational Intelligence and Knowledge Economy (ICCIKE), Dubai, United Arab Emirates, 11\u0026ndash;12 December 2019; pp. 334\u0026ndash;339. \u003c/li\u003e\n\u003cli\u003eQaddoura, R.; Biltawi, M.M. Improving Fraud Detection in An Imbalanced Class Distribution Using Different Oversampling Techniques. In Proceedings of the 2022 International Engineering Conference on Electrical, Energy, and Artificial Intelligence (EICEEAI), Zarqa, Jordan, 29 November\u0026ndash;1 December 2022; pp. 1\u0026ndash;5. \u003c/li\u003e\n\u003cli\u003eSailusha, R.; Gnaneswar, V.; Ramesh, R.; Rao, G.R. Credit Card Fraud Detection Using Machine Learning. In Proceedings of the 2020 4th International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 13\u0026ndash;15 May 2020; pp. 1264\u0026ndash;1270.\u003c/li\u003e\n\u003cli\u003eJain, Y.; Tiwari, N.; Dubey, S.; Jain, S. A comparative analysis of various credit card fraud detection techniques. \u003cem\u003eInt. J. Recent Technol. Eng.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e7\u003c/em\u003e, 402\u0026ndash;407. \u003c/li\u003e\n\u003cli\u003ePrasad, P.Y.; Chowdary, A.S.; Bavitha, C.; Mounisha, E.; Reethika, C. A Comparison Study of Fraud Detection in Usage of Credit Cards using Machine Learning. In Proceedings of the 2023 7th International Conference on Trends in Electronics and Informatics (ICOEI), Tirunelveli, India, 11\u0026ndash;13 April 2023; pp. 1204\u0026ndash;1209. \u003c/li\u003e\n\u003cli\u003eSadgali, I.; Sael, N.; Benabbou, F. Performance of machine learning techniques in the detection of financial frauds. \u003cem\u003eProcedia Comput. Sci.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e148\u003c/em\u003e, 45\u0026ndash;54. \u003c/li\u003e\n\u003cli\u003eSaputra, A.; Suharjito. Fraud detection using machine learning in e-commerce. \u003cem\u003eInt. J. Adv. Comput. Sci. Appl.\u003c/em\u003e\u003cstrong\u003e2019\u003c/strong\u003e, \u003cem\u003e10\u003c/em\u003e, 332\u0026ndash;339. \u003c/li\u003e\n\u003cli\u003eForough, J.; Momtazi, S. Ensemble of deep sequential models for credit card fraud detection. \u003cem\u003eAppl. Soft Comput.\u003c/em\u003e\u003cstrong\u003e2021\u003c/strong\u003e, (\u003cem\u003e99)\u003c/em\u003e, 106883. \u003c/li\u003e\n\u003cli\u003eKarthik, V.S.S.; Mishra, A.; Reddy, U.S. Credit card fraud detection by modelling behaviour pattern using hybrid ensemble model. \u003cem\u003eArab. J. Sci. Eng.\u003c/em\u003e\u003cstrong\u003e2022\u003c/strong\u003e, \u003cem\u003e47(1)\u003c/em\u003e, 1987\u0026ndash;1997.\u003c/li\u003e\n\u003cli\u003eSonam Gupta, A Hybrid Machine Learning Approach for Credit Card Fraud detection, India International Journal of Information Technology Project Management, Volume 13 Issue 3, pp. 1-13, November 2022.\u003c/li\u003e\n\u003cli\u003eSayali Saraf1, Anupama Phakatkar, Detection of Credit Card Fraud using a Hybrid Ensemble Model, International Journal of Advanced Computer Science and Applications, Vol. 13, No. 9, pp 464-474, 2022.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Credit card fraud detection, hybrid ensemble approach, ensemble learning, machine learning, data imbalance, enhanced (augmented) dataset, and disjunctive voting ensemble classification mechanism","lastPublishedDoi":"10.21203/rs.3.rs-7237183/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7237183/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eCredit card fraud detection remains a critical yet challenging task due to the extreme class imbalance inherent in transaction datasets, where fraudulent activities constitute only a small fraction of the total records. To address this imbalance and enhance the detection of rare fraud instances, this study proposes a novel hybrid framework that integrates density-based clustering for data augmentation with an ensemble classification strategy optimized for high recall. In the preprocessing stage, the proposed method utilizes Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to identify minority-class clusters and synthetically augment the fraud class. This step aims to preserve the intrinsic structure of fraudulent patterns while increasing their representation in the training set. Subsequently, an ensemble model comprising Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM) classifiers is constructed. Final predictions are generated using a Disjunctive Voting Ensemble (DVE) strategy, wherein a transaction is classified as fraudulent if any of the base classifiers predicts it as such. This permissive voting mechanism prioritizes recall, thereby minimizing the risk of undetected fraudulent transactions. Extensive experiments conducted on three publicly available credit card fraud imbalanced datasets containing transaction records from European credit cardholders in 2023, providing a realistic fraud detection scenario. Computational efficiency in both training and testing is achieved using the Anaconda Navigator (Spider-Python 3.12) environment. The comparative analysis shows that the proposed DBSCAN-augmented DVE framework delivers notable improvements over traditional ensemble approaches and single-model baselines, particularly in recall and F1-score, while preserving consistently high precision. This approach proves to be both robust and interpretable, making it highly applicable to real-world fraud detection scenarios characterized by severe class imbalance. The results are especially compelling, achieving recall and F1-scores as high as 99.5% and 99.8% respectively, and consistently maintaining a perfect accuracy and precision of 100% across all imbalanced datasets. This study highlights the effectiveness of hybrid ensemble approaches in combating credit card fraud. The findings lay the groundwork for developing more resilient and adaptive fraud detection systems, which are crucial in countering the evolving tactics of fraudsters. The proposed model marks a significant advancement in securing financial transactions and mitigating risks in an increasingly digital economy.\u003c/p\u003e","manuscriptTitle":"Enhancing Credit Card Fraud Detection Using DBSCAN-Augmented Disjunctive Voting Ensemble","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-26 06:36:13","doi":"10.21203/rs.3.rs-7237183/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-02T13:28:07+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-23T19:23:50+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"34113462102797441020892850966893505374","date":"2025-08-23T15:47:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-22T16:19:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"54103737122721341790230780664784368623","date":"2025-08-22T15:58:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-17T08:00:07+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-08-05T06:21:42+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-01T09:29:26+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-07-31T08:46:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-07-28T20:49:24+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"3fc35eb6-1756-4ae2-a122-3332dc34ec8f","owner":[],"postedDate":"August 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":53612997,"name":"Physical sciences/Engineering"},{"id":53612998,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2025-11-17T16:03:16+00:00","versionOfRecord":{"articleIdentity":"rs-7237183","link":"https://doi.org/10.1038/s41598-025-22960-w","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-11-13 15:58:22","publishedOnDateReadable":"November 13th, 2025"},"versionCreatedAt":"2025-08-26 06:36:13","video":"","vorDoi":"10.1038/s41598-025-22960-w","vorDoiUrl":"https://doi.org/10.1038/s41598-025-22960-w","workflowStages":[]},"version":"v1","identity":"rs-7237183","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7237183","identity":"rs-7237183","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}