Fraud Detection in the Banking Sector using Machine Learning

Project Overview

This project focuses on detecting fraudulent transactions in the banking sector using machine learning techniques. Fraud detection is a critical challenge because fraudulent transactions are rare compared to legitimate ones, leading to imbalanced datasets.

To address this, I applied multiple ML models and used SMOTE (Synthetic Minority Oversampling Technique) to handle the imbalance.

Models Used

• 🌳 Random Forest Classifier
• ⚡ XGBoost Classifier
• 📈 Support Vector Machine (SVM with RBF kernel)

Results Summary

Model	Accuracy	Precision (Fraud)	Recall (Fraud)	F1-Score (Fraud)
Random Forest + SMOTE	99.95%	84.3%	87.7%	86.0%
XGBoost + SMOTE	99.94%	83.0%	84.7%	83.8%
SVM + RobustScaler + SMOTE	99.75%	39.1%	85.7%	53.7%

Best Model: Random Forest (highest balance of recall and F1-score).
Key Insights

• Accuracy alone is not enough for imbalanced problems like fraud detection.
• Recall and F1-score are more important, since missing fraud is riskier than false alarms.
• Random Forest performed best, while SVM produced more false positives.

Tech Stack

• Python 🐍
• Jupyter Notebook
• Libraries: pandas, numpy, scikit-learn, xgboost, imbalanced-learn, matplotlib

Project Structure

fraud-detection-ml working_with_XGBOOST_RF_SVM.ipynb # Jupyter Notebook with full code README.md # Project documentation results # Screenshots of confusion matrices, reports, charts

Dataset

• Credit Card Fraud Detection Dataset from Kaggle.
• Contains 284,807 transactions with 492 frauds (0.17%).

Sample Output

• Confusion Matrices for each model
• Classification Reports
• Comparison Chart (Recall vs F1-Score)

Acknowledgments

• Dataset provided by Kaggle.
• Libraries: scikit-learn, imbalanced-learn, XGBoost.

Author: Daniel Udeme
*Email: xavidanito35@gmail.com