Embedded Files
Why Accurate Fraud Detection Isn't Enough
Financial Scale
Financial Scale
Credit card fraud causes billions in annual global losses. Automated detection is no longer optional. It's infrastructure.
Regulatory Pressure
Regulatory Pressure
EU AI Act Articles 10 & 13 classify automated fraud detection as high-risk AI, mandating individualised explanation records for every decision.
The Trust Deficit
The Trust Deficit
Only 32% of consumers trust AI systems with sensitive payment data. Black-box models are a legal AND reputational liability.
The Fundamental Conflict
The Fundamental Conflict
Existing fraud detection models optimized purely for predictive performance systematically fail to produce the audit-ready documentation required under emerging financial laws. The models that detect fraud best are also the models regulators cannot approve. This research directly addresses that conflict.
System Assumptions
System Assumptions
Transaction data in standard tabular format
Historical patterns reflect European cardholder behaviors
Interpretation tools applied after training only
Critical Constraints
Critical Constraints
PCA anonymization — no original feature names available
Extreme 578:1 class imbalance requiring synthetic data
False negatives = direct financial loss; false positives = customer friction
Page updated
Report abuse