Upload any binary classification CSV and get fraud predictions, SHAP explainability, and cost-optimized thresholds — no code required.
Built for analysts, researchers, and students who want production ML power without writing a single line of code.
Automatically selects and tunes the best model for your dataset. Supports 100 rows all the way up to 2 million rows across 6 smart size tiers.
Every prediction comes with a full SHAP breakdown — see exactly which features drove the fraud decision and by how much.
Balances false positives vs false negatives based on your real business cost model — not just raw accuracy metrics.
Upload a CSV with thousands of transactions. Get risk scores, confidence intervals, and downloadable results instantly.
Monitors your model for data drift and performance degradation over time — alerts you before accuracy degrades in production.
Full model versioning with MLflow. Champion/challenger promotion, experiment tracking, and fully reproducible training runs.
No ML knowledge required. Just bring your data.
Drop any binary classification dataset. AutoML-X detects columns, handles missing values, and engineers features automatically — no preprocessing needed.
The engine trains, evaluates, and selects the best model. You get fraud probability scores with confidence intervals for every single row.
Explore SHAP explanations per prediction, tune your business decision threshold, and download results — all in a clean, interactive dashboard.
Validated on real transaction data with SMOTE-balanced training and cost-optimized threshold selection.
Built with battle-tested open source tools. Dockerized and deployed on HuggingFace Spaces with full CI/CD automation.
Free, open-source, and running live right now. No signup required.