Skip to content

README.md

Latest commit

 

History

History
89 lines (62 loc) · 3.21 KB

README.md

File metadata and controls

89 lines (62 loc) · 3.21 KB

💳 Credit Card Fraud Detection

This project demonstrates Credit Card Fraud Detection using a publicly available dataset, showcasing the complete pipeline from data exploration to model deployment.

📌 Project Overview

The goal of this project is to build a model that can accurately identify fraudulent credit card transactions.
We use the Credit Card Fraud Detection dataset and employ a Logistic Regression model, while addressing the significant class imbalance present in the data.

📂 Dataset

  • Dataset: Credit Card Fraud Detection (via fetch_openml)
  • Features:
    • V1–V28 → anonymized transaction features (PCA transformed)
    • Amount → transaction amount
    • Class → target variable (0 = non-fraud, 1 = fraud)

🛠 Steps Taken

1️⃣ Data Loading

  • Loaded into a pandas DataFrame for analysis.

2️⃣ Data Exploration

  • Checked data structure, missing values, and target variable imbalance.

3️⃣ Data Preprocessing

  • Applied StandardScaler for feature scaling.
  • Split into train/test sets with stratification to preserve class ratios.
  • Applied SMOTE (Synthetic Minority Over-sampling Technique) to handle severe class imbalance.

4️⃣ Model Training

  • Trained a Logistic Regression model on the SMOTE-resampled training set.

5️⃣ Model Evaluation

  • Evaluated on the unseen test set using multiple metrics:
    • ✅ Precision
    • ✅ Recall
    • ✅ F1-Score
    • ✅ Confusion Matrix
    • ✅ ROC AUC Score

6️⃣ Model Deployment

  • Saved the trained model for future predictions on new transactions.

📊 Results

Key findings from model evaluation:

  • Recall (Fraudulent Class): 0.92 ✅ (captures most fraud cases)
  • Precision (Fraudulent Class): 0.06 ⚠️ (high false positives)
  • ROC AUC Score: 0.9707 🚀 (strong discriminative ability)

🔎 Interpretation:
The model is excellent at catching frauds (high recall), but suffers from low precision (too many false alarms). Further optimization (e.g., advanced models, threshold tuning, ensemble methods) is needed.

📦 Tech Stack

  • Python 🐍
  • Pandas, NumPy → Data handling
  • Scikit-learn → Preprocessing, modeling, evaluation
  • Imbalanced-learn (SMOTE) → Handling class imbalance

🚀 Future Improvements

  • Experiment with tree-based models (Random Forest, XGBoost, LightGBM).
  • Apply threshold tuning for better precision-recall tradeoff.
  • Use anomaly detection methods for fraud detection.
  • Build a real-time detection system with streaming data.

📑 References

🏁 Conclusion

This project provides a solid baseline for fraud detection with Logistic Regression and SMOTE.
It highlights the challenges of imbalanced data and the trade-off between recall vs. precision in financial fraud detection.