Fraud Detection - Credit Card Transactions

Personal project carried out in September 2021. Inspired by real-life finance use cases, this project aims to help a company automatically detect fraud in banking transactions, in order to protect customers against unauthorized payments.

❓ Issue

With millions of transactions processed every day, how to identify quickly and reliably the suspicious operations?
➡️ The goal is to build a robust classification model capable of differentiating normal transactions from fraud, based on an annotated dataset.

🛠️ Implemented solution

• User management: registration, login, account subscriptions

• 📊 Exploratory data analysis to understand distribution, class imbalance, and key variables.

• 🔍 Implementation and comparison of several supervised classification models:

  • XGBoost

  • Random Forest

  • SVM

  • K-Nearest Neighbors

  • Logistic Regression

• ⚖️ Use of imbalanced data treatment techniques

• 🧮 Evaluation through appropriate metrics: precision, recall, F1-score, confusion matrix.

• 💡 Some forecasting attempts on temporal trends.

⚙️ Technical stack

• Language: Python

• Libraries: pandas · seaborn · scikit-learn · xgboost

• Methods: classification · model evaluation · management of imbalanced classes

• Environment: Jupyter Notebook

• Prerequisites: Machine Learning · Python · basic statistics