A Comparative Study of Random Forest and XGBoost for Detecting Credit Card Fraud Transactions using Big Data | Mohamed Riham - CRP Final Presentation.pptx

The presentation titled "Mohamed Riham - CRP Final Presentation.pptx" details "A Comparative Study of Random Forest and XGBoost for Detecting Credit Card Fraud Transactions using Big Data" , a project aimed at finding the most efficient detection mechanism for credit card fraud. The core of the study involved evaluating and comparing two machine learning algorithms, Random Forest (an ensemble of decision trees) and XGBoost (a Gradient Boosting method) , using a dataset of over 1 million transactions. To handle the significant class imbalance (99.48% non-fraud vs. 0.52% fraud) , the SMOTE technique was applied. Performance was measured across multiple metrics, including accuracy, precision, recall, F1-score, and AUC-ROC. The conclusion was definitive: XGBoost outperformed Random Forest in all metrics , a finding supported by the rejection of the null hypothesis with a P-Value of 0.0003. The study deemed XGBoost, which was particularly strong in precision and recall , to be ideal for fraud detection in imbalanced datasets. Future work recommended is to incorporate customer behavior, explore hybrid models, and test a real-time implementation.

1. A Comparative Study of Random Forest and XGBoost for Detecting Credit Card Fraud Transactions using Big Data Learner Name – M.A. Mohamed Riham Student Id - 1028401 HND in Computing 21 Batch Assessor : A.L.F.Sajeetha Supervisor: Mr. Mohamed Nismy

2. 4/21/2025 2 CRP Abstract Credit card fraud is a critical issue requiring efficient detection mechanisms. Compared Random Forest and XGBoost using over 1 million transactions. Applied SMOTE to handle class imbalance. XGBoost outperformed Random Forest in all metrics. Keywords: Credit Card Fraud, Random Forest, XGBoost, SMOTE, Big Data.

3. Research Objectives 4/21/2025 3 CRP

4. 4/21/2025 4 CRP Research Objectives Evaluate Random Forest and XGBoost for fraud detection. Compare performance: accuracy, precision, recall, F1-score, AUC-ROC. Identify the best-suited algorithm for real-world use.

5. 3. Research Question 4/21/2025 5 CRP

6. How do Random Forest and XGBoost compare in detecting fraudulent credit card transactions? Dependent variable: Fraud detection accuracy, precision, recall, F1-score, and AUC-ROC Independent variable: Transaction features (amount, location, merchant category) 4/21/2025 6 CRP Research Question

7. 7. Methodology 4/21/2025 7 CRP

8. Methodology 4/21/2025 8 CRP Models: Random Forest XGBoost Tools: Python, Scikit-learn, TensorFlow, Pandas, NumPy, Matplotlib Evaluation Metrics: Accuracy, Precision, Recall, F1-score, AUC-ROC

9. 5. Model Explanation 4/21/2025 9 CRP

10. Model Explanation Random Forest Ensemble of decision trees. Reduces overfitting via random sampling. XGBoost Gradient Boosting. Sequentially corrects previous errors. 4/21/2025 13 CRP

11. 6.Data Visualization 4/21/2025 11 CRP

12. Data Visualization 4/21/2025 12 CRP

13. Model Performance Metrics 4/21/2025 13 CRP

14. 4/21/2025 CRP 14 Metric Random Forest XGBoost Accuracy 99.63% 99.74% Precision 99.66% 99.71% Recall 99.63% 99.74% F1-Score 99.65% 99.72%

15. Hypothesis Testing 4/21/2025 15 CRP

16. Hypothesis Testing 4/21/2025 16 CRP Null Hypothesis (H0) : No performance difference. Alternate Hypotheses (H1) : XGBoost outperforms Random Forest. P – Value: 0.0003 → Reject H0 XGBoost shows significantly better performance

17. Findings 4/21/2025 17 CRP

18. Findings 4/21/2025 18 CRP XGBoost consistently outperformed Random Forest. Particularly strong in precision and recall. Ideal for fraud detection in imbalanced datasets.

19. Limitations & Future work 4/21/2025 19 CRP

20. Limitation and Future Work 4/21/2025 20 CRP No real-time implementation tested. Future work: Incorporate customer behavior. Test in real-time settings. Explore hybrid/ensemble models.

21. Thank you 4/21/2025 21 CRP