A neural network has learned to discover its own fraud detection rules without human intervention, marking a breakthrough in neuro-symbolic AI. The experiment, conducted on the Kaggle Credit Card Fraud dataset with a 0.17% fraud rate, achieved ROC-AUC 0.933 ± 0.029 while maintaining 99.3% fidelity to the neural network's predictions.

Unlike traditional systems where humans write rules, this hybrid model automatically extracts IF-THEN fraud rules during training. The model independently rediscovered V14, a feature long known by analysts to correlate strongly with fraud, without being told to look for it. This demonstrates how neural networks can combine statistical learning with human-readable logic, creating interpretable rules that compliance teams can audit and defend.

The architecture uses a differentiable rule-learning module that learns to express the neural network's decisions as symbolic rules. The model learned to weight the neural path at roughly 88% and the rule path at 12% on average. This approach addresses a critical gap in fraud detection where teams need both the performance of machine learning and the interpretability required for regulatory compliance.