title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned
LeakGuard	🛡️	purple	blue	gradio	6.5.0	app.py	false

LeakGuard

A web app that analyzes a CSV dataset BEFORE model training and detects silent data leakage risks that commonly cause models to fail in production.

What It Detects

Type	Detection Method	Risk Indicators
Target Leakage	Mutual Information, Pearson & Spearman correlation	Features containing direct/indirect target information
Time Leakage	Correlation drift, rolling window analysis	Future information leaking into past samples
Duplicate Leakage	Row hashing, entity ID overlap	Same samples appearing across splits
Proxy Leakage	Feature importance instability	Hidden proxies acting as target substitutes

Quick Start

1. Upload your CSV dataset
2. Select target column (required)
3. Select time & entity ID columns (optional)
4. Click Analyze to get instant results

What You Get

• Feature Risk Table - Detailed risk assessment with MI, Pearson, Spearman scores
• Visual Analytics - 5 interactive charts showing leakage patterns
• Risk Summary - Overall leakage risk across all categories

Tech Stack

• Frontend: Gradio
• Data Processing: Pandas, NumPy
• ML Detection: Scikit-learn (Random Forest)
• Statistics: SciPy (Spearman, MI)
• Visualization: Matplotlib

Features

✅ CPU-only (no GPU required)
✅ Explainable results with statistical basis
✅ Fast analysis (seconds for typical datasets)
✅ Production-ready architecture

Links

• Live Demo: Hugging Face Space
• GitHub: Source Code

📝 License

Apache 2.0

---

Built for Kaggle & Hugging Face Spaces | © 2026