Dataset and benchmark measurements for: Quantifying Layered Security Overhead in Explainable ML Fraud Detection Systems

Synthetic labelled anomaly dataset (500 records, 5 features, 3 injected anomaly patterns), 70/30 train-test split indices, One-Class SVM normal-only training subset, per-layer benchmark measurements (6 security configurations × 3 runs × 4 endpoints), per-model evaluation metrics (Isolation Forest, Local Outlier Factor, One-Class SVM), and all result figures for the accompanying paper. Dataset generated via Geometric Brownian Motion with fixed seeds for full reproducibility.