Banking fraud prevention with explainable risk scoring

Banking fraud prevention with explainable risk scoring is an applied project that develops an interpretable framework for fraud-risk scoring from transactional networks. The goal is to help risk and operations teams prioritize review, monitoring, and follow-up using relational signals rather than relying only on black-box predictions.

The work connects CRiSS-LAB expertise in network science, graph analytics, machine learning, anomaly detection, and decision support. At a high level, the approach studies how accounts interact within local transaction neighborhoods, how exposure to previously flagged risk propagates through the network, and which structural patterns suggest concentration, dispersion, rapid movement, recency, or unusual neighborhood behavior.

The project is intentionally documented here only at a public and non-sensitive level. It does not disclose private data, operational rules, third-party-specific materials, or implementation details that could compromise security or fraud-prevention workflows.

Public-facing research value:

• Explainable risk scoring for high-stakes decisions.
• Graph-based features for transactional systems.
• Human-in-the-loop prioritization for risk teams.
• Responsible AI practices for fraud detection and financial operations.
• Better separation between prediction, interpretation, and decision-making.