Acquiring fraud detection with methodological discipline, economic realism, and governance by design.

What this hub can provide

• Time aware evaluation protocol, rolling windows, leakage controls, and reproducible reporting.
• Cost model and threshold policy based on expected net benefit under operational constraints.
• Benchmarking logic where strong baselines come first and deep learning is justified only when it wins under temporal backtests.
• Governance pack with model cards, release gates, monitoring plan, rollback triggers, and audit evidence.

Practical scope. Merchant acquiring, card present and e-commerce, designed for regulated environments.

Use modes

• Research facing material for reviewers, supervisors, and academic collaborators.
• Method facing material for fraud teams that need defensible evaluation and threshold policy.
• Translation layer between academic standards and real world fraud operations.
• Public reference point for protocols, templates, and decision logic that can be scrutinised without disclosing sensitive data.

The site is not a rule repository and not a vulnerability guide. It exists to document method, evidence, and limits.

Credibility signals

Research questions

• How can false positives be reduced without sacrificing fraud capture under strict temporal validation?
• How should fraud models be evaluated when the target is expected net benefit rather than generic accuracy?
• How can drift be detected and controlled in acquiring, where merchant behaviour and attacker behaviour evolve continuously?
• How can governance, auditability, and human oversight be operationalised without undermining deployment realism?

Scope. Merchant context, behavioural signals, cost aware decision policies, and deployment compatible evaluation.

Facts and figures

Core outputs on this hub

Academic supervision and research grounding

CyberAntifraud is grounded in doctoral research conducted at NOVA IMS, Universidade Nova de Lisboa, under the academic supervision of Professor Vítor Santos. This foundation reinforces the methodological orientation of the project, including clear problem framing, reproducible evaluation, time aware validation, and explicit attention to operational and governance constraints.

The site is authored and maintained by Paulo Saramago as an independent public research hub. Its purpose is to translate doctoral work into structured public artefacts that are scientifically credible, operationally relevant, and suitable for discussion with both academic and industry audiences.

Supervisor profile

Professor: Vítor Santos

Role: Associate Professor

Institution: NOVA IMS, Universidade Nova de Lisboa

Google Scholar · NOVA IMS profile

Positioning statement

CyberAntifraud is positioned as an applied research and methods hub for acquiring fraud detection. It is not a generic AI commentary site. Its emphasis is on methodological clarity, temporal validity, cost sensitive evaluation, reproducibility, and governance readiness in regulated decision environments.

Evaluation protocol

Blocked time series validation, rolling windows, leakage control, stability tracking, and release gates designed to approximate production reality.

PDF: Download

Cost model note

False positives and false negatives translated into comparable economic terms so threshold decisions can be challenged, defended, and governed.

PDF: Download

Model card template

Audit ready documentation for each model release, including data windows, metrics, drift checks, rollback criteria, and sign off evidence.

PDF: Download

Why these artefacts matter

The protocol pack exists to make claims defensible. The goal is not only to improve metrics, but to make evaluation honest, threshold choices explicit, and deployment decisions compatible with audit and operational reality.