Fraud Detection Using XGBoost

Problem Statement

Build a machine learning model to detect fraudulent financial transactions and provide business insights.

Dataset

• 6.3M transactions
• Highly imbalanced dataset
• Columns include transaction type, amount, balances

Note: Dataset not uploaded due to large size.

Approach

• Data cleaning & feature engineering
• Label encoding for categorical variables
• Time-based train-test split
• XGBoost model for fraud detection

Model Performance

• Precision (Fraud): 0.93
• Recall (Fraud): 0.89
• F1-score: 0.91

Tools Used

• Python
• Pandas, NumPy
• XGBoost
• Scikit-learn
• Matplotlib

Key Learnings

• Fraud detection requires imbalance-aware metrics
• Tree-based models do not require feature scaling
• Business understanding is critical in feature engineering