A newer version of the Streamlit SDK is available:
1.52.1
metadata
sdk: streamlit
sdk_version: 1.51.0
license: apache-2.0
💳 Credit Card Fraud Detection Dashboard
📌 Overview
Interactive dashboard built with Streamlit, Plotly, and Scikit-learn for real-time fraud detection analysis.
It demonstrates a business-aware ML pipeline on the classic Credit Card Fraud Dataset (284,807 transactions, only 492 frauds ≈ 0.17%).
- 🔍 Upload your own transaction CSV or use the built-in dataset
- ⚖️ Custom decision thresholds with cost-sensitive analysis
- 📊 Confusion matrix, ROC/PR curves, and cost–threshold visualization
- 💡 Permutation feature importance for interpretability
- 🧾 Segmented performance profiling (by amount, time of day, etc.)
📊 Dashboard Preview
Data Overview
Prediction Engine
Model Metrics
Model Insights
Data Quality & Segments
🔑 Features
- Models: RandomForest & XGBoost (calibrated)
- Presets: Strict / Balanced / Lenient thresholds
- Threshold Finder: auto-select by target Precision/Recall
- Cost Analysis: business-aligned FP vs FN costs
- Visuals: Confusion matrix, ROC, PR, cost vs threshold curves
- Insights: Permutation importance, segmented KPIs
- Data Handling: automatic schema validation + engineered features (
log(Amount), business hours, night proxy)
🚀 Run Locally
Clone the repo and install requirements:
git clone https://github.com/tarekmasryo/fraud-detection-dashboard.git
cd fraud-detection-dashboard
pip install -r requirements.txt
Run the app:
streamlit run app.py
☁️ Deploy on Hugging Face Spaces
This repository is ready to be deployed as a Streamlit Space on Hugging Face.
Make sure to include the following files in your repo:
app.py→ main app filerequirements.txt→ Python dependenciesartifacts/→ trained model.joblibfiles andthresholds.jsondata/creditcard.csv(optional, for default dataset)
📜 License & Attribution
- Data → Original Credit Card Fraud Dataset
Licensed under CC BY-NC 4.0 — for research & educational use only.