🛡️ dataleak-guard

The Fast, Automated Machine Learning Data Leakage & Train-Test Audit Toolkit.
Catch train-test contamination, proxy target leaks, ID memorization, and distribution drift before training breaks production.

📌 Why dataleak-guard?

In Machine Learning and Data Science, Data Leakage is the #1 silent killer of predictive models:

• A model achieves 99.5% accuracy on your Jupyter notebook.
• In production, performance collapses to random chance (52%).
• The culprit? Subtle row overlaps, future variables collected post-event, or improper train-test preprocessing splits.

dataleak-guard audits your train and test datasets in under 1 second, computes a standardized Leakage Risk Score (0–100), and provides actionable remediation fixes.

🚀 Key Features

📦 Installation

pip install dataleak-guard

⚡ Quickstart (Python API)

import dataleak_guard as dlg
from sklearn.model_selection import train_test_split

# Split your data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Audit in 1 line
report = dlg.audit(X_train, X_test, y_train, y_test)

# Display colored terminal dashboard
report.show()

# Export report to Markdown or JSON
print(report.to_markdown())
json_report = report.to_json()

🖥️ Command Line Interface (CLI)

Audit datasets directly from your terminal:

# Audit CSV files
dataleak-guard audit train.csv test.csv --target churn

# Export audit to Markdown report
dataleak-guard audit train.csv test.csv --target churn --output audit_report.md

# Run the live interactive demo
dataleak-guard demo

📊 Terminal Dashboard Preview

╭────────────────────────────────────────────────────────────────────╮
│ 🛡️  DATALEAK-GUARD: ML Data Leakage Audit Report                  │
│ Train samples: 140 | Features: 5    Test samples: 65 | Features: 5 │
│ Target column: churn                                               │
│ Leakage Risk Score: 88/100 — Status: CRITICAL LEAKAGE DETECTED     │
╰────────────────────────────────────────────────────────────────────╯
                  Detected Leakage Findings & Vulnerabilities                   
┏━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ S… ┃ Category ┃ Issue Title      ┃ Affected     ┃ Actionable Fix             ┃
┡━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ 🚨 │ Row      │ Train-Test Row   │ -            │ Deduplicate dataset before │
│ C… │ Contami… │ Overlap Detected │              │ split or use GroupKFold    │
├────┼──────────┼──────────────────┼──────────────┼────────────────────────────┤
│ 🚨 │ Target   │ Extreme          │ account_clo… │ Drop variables that cannot │
│ C… │ Leakage  │ Feature-Target   │              │ be observed at inference   │
│    │          │ Correlation      │              │ time                       │
├────┼──────────┼──────────────────┼──────────────┼────────────────────────────┤
│ ⚠️ │ Feature  │ Identifier       │ customer_id  │ Remove identifier columns  │
│ W… │ Enginee… │ Columns          │              │ before model training      │
└────┴──────────┴──────────────────┴──────────────┴────────────────────────────┘

🤖 CI/CD Integration (GitHub Actions)

Fail your machine learning training pipeline if data leakage risk exceeds acceptable thresholds:

import sys
import dataleak_guard as dlg

report = dlg.audit(X_train, X_test, y_train, y_test)
if report.risk_score > 30:
    print(f"FAILED: Data leakage score {report.risk_score}/100 exceeds safety threshold!")
    sys.exit(1)

🛠️ Development & Testing

# Clone the repository
git clone https://github.com/sumit-2007-git/leakdetect.git
cd leakdetect

# Install in editable mode
pip install -e .

# Run test suite
pytest tests -v

# Run linter
ruff check dataleak_guard tests

📄 License

Distributed under the MIT License. See LICENSE for more information.

👤 Author

• Sumit — GitHub (@sumit-2007-git)