⚡ GPU-Accelerated Fraud Detection Demo

A live, runnable benchmark showing CPU vs GPU inference speed for financial fraud detection using XGBoost and the PaySim dataset (6.36M transactions).

Click the badge above → change runtime to T4 GPU → Run All. Done in ~10 minutes.

What you'll see

• Real transaction data loaded from Hugging Face (no Kaggle login, no manual downloads)
• XGBoost fraud model trained and evaluated (AUC-ROC > 0.99)
• Side-by-side throughput and latency benchmark: CPU vs GPU
• Cost-per-million-inference estimate using AWS on-demand pricing

Why this matters in BFSI

Real-time card authorisation has a ~200ms window. At peak volumes — festive sales, IPO days — CPU-based inference creates queues. GPU inference absorbs the burst within SLA and costs less per inference at scale.

---