Credit Card Fraud Detection Using Neural Networks
Credit card fraud detection with MLP neural networks and feature engineering for imbalanced data.
Scientific Landing Context
This page presents a scientific synthesis of "Credit Card Fraud Detection with Neural Networks", structured for academic reading, methodological auditing, and DOI-ready preparation.
Fraude financeira combina alta assimetria de classes com necessidade de baixa latencia decisoria em tempo quase real. Pergunta de pesquisa: Como a abordagem proposta em "Detecção de Fraudes em Cartões com Redes Neurais" pode reduzir risco sistemico e ampliar confiabilidade decisoria em ambiente real?
- Estrutura de avaliacao orientada a risco economico de fraude.
- Integração de calibracao de probabilidade com politicas operacionais.
- Boas praticas para monitorar drift em cenarios de pagamento digital.
Suporte a motores antifraude em emissores, adquirentes e fintechs com trilha explicavel para auditoria. The full version includes implications for engineering, governance, and reproducibility.
The complete PDF features a formal scientific structure (Abstract, Introduction, Development, Final Considerations, and References), with bibliography verifiable by URL/DOI.
Abstract — Portuguese
Card fraud detection with MLP neural networks and feature engineering for imbalanced data. The central problem investigated is: Financial fraud combines high class asymmetry with the need for low decision latency in near real-time. A methodological design was adopted focusing on internal validity, comparability, and reproducibility: Supervised pipeline with resampling, threshold calibration, and evaluation by precision-recall and error cost. The main results indicate that the combination of MLP with threshold adjustment improves fraud capture while maintaining an acceptable operational false positive rate. The methodological contribution includes an audit-oriented scientific writing standard, with premise tracking, boundary delimitation, and explicit connection between theory and implementation implications. The objective of this work is to structuredly evaluate how "Card Fraud Detection with Neural Networks" can generate scientific and operational value with methodological traceability. In summary, the study offers a technical basis for decision-making with verifiable bibliography and guidance for a DOI-ready version. (Ngai, 2011).
Abstract — English
This article presents a reproducible, high-rigor synthesis of "Detecção de Fraudes em Cartões com Redes Neurais" by aligning methodological traceability, interdisciplinary evidence, and operational recommendations for deployment contexts with explicit governance constraints. (Whitrow, 2009).
Introduction
In the current state of the art, financial fraud combines high class asymmetry with the need for low decision latency in near real-time. Card fraud detection with MLP neural networks and feature engineering for imbalanced data. (Jurgovsky, 2018).
The research gap lies in the absence of integration between theoretical formulation, operational criteria, and transparent validation mechanisms. The objective of this work is to structuredly evaluate how "Card Fraud Detection with Neural Networks" can generate scientific and operational value with methodological traceability. (Carcillo, 2021).
Research question: How can the approach proposed in "Card Fraud Detection with Neural Networks" reduce systemic risk and enhance decision reliability in a real environment? The relevance of the study stems from its potential for application in highly critical scenarios, where predictability, security, and decision quality are mandatory requirements. (Bahnsen, 2016).
Methodology
Methodological design: Supervised pipeline with resampling, threshold calibration, and evaluation by precision-recall and error cost. The protocol prioritizes premise traceability, explicit scope delimitation, and comparison between technical alternatives. (Whitrow, 2009).
The analytical strategy combines bibliographic triangulation, internal consistency criteria, and evidence-oriented reading. Where applicable, the study adopts controls to reduce selection biases, information leakage, and non-reproducible conclusions. (Jurgovsky, 2018).
For reliability, checkpoints were defined at each stage: problem definition, argumentative construction, results confrontation, and consolidation of practical implications. (Carcillo, 2021).
Development and Results
Main result: The combination of MLP with threshold adjustment improves fraud capture while maintaining an acceptable operational false positive rate. (Ngai, 2011).
Direct contributions: Evaluation framework oriented to the economic risk of fraud. Integration of probability calibration with operational policies. Best practices for monitoring drift in digital payment scenarios. (Whitrow, 2009).
Performance depends on continuous updates and behavioral drift governance. The interpretation of results was carried out in contrast with primary literature and with an emphasis on coherence between theory, method, and application. (NIST, 2026).
From an applied perspective, the findings indicate that evidence-based structuring improves decision clarity, reduces implementation ambiguity, and strengthens technical governance for production operation. (Jurgovsky, 2018).
Limitations: The generalization of the findings depends on replication in additional samples, with different data regimes and temporal horizons. The availability of data with adequate granularity may limit comparability between distinct institutional environments. (Ngai, 2011).
Discussion
Recommendations
- Evaluation framework oriented to the economic risk of fraud. (Jurgovsky, 2018).
- Integration of probability calibration with operational policies. (Carcillo, 2021).
- Best practices for monitoring drift in digital payment scenarios. (Bahnsen, 2016).
- Replicate the study in new operational contexts with a quasi-experimental design. (NIST, 2026).
- Deepen metrics of robustness, explainability, and economic impact under uncertainty. (Ngai, 2011).
Conclusion
Support for anti-fraud engines in issuers, acquirers, and fintechs with an explainable audit trail. The study delivers a scientific artifact with a structure ready for indexing, citation, and future DOI assignment. (Bahnsen, 2016).
Continuity agenda: Replicate the study in new operational contexts with a quasi-experimental design. Deepen metrics of robustness, explainability, and economic impact under uncertainty. Prepare a DOI-ready version with a data package, protocol, and methodological appendix. (NIST, 2026).
References (Harvard)
- Ngai, E. W. T. et al. (2011). The application of data mining techniques in financial fraud detection. Source
- Whitrow, C. et al. (2009). Transaction aggregation as a strategy for credit card fraud detection. Source
- Jurgovsky, J. et al. (2018). Sequence classification for credit-card fraud detection. Source
- Carcillo, F. et al. (2021). Combining unsupervised and supervised learning in credit card fraud detection. Source
- Bahnsen, A. C. et al. (2016). Classifying highly imbalanced data using cost-sensitive decision trees. Source
- NIST. AI Risk Management Framework 1.0. Source
How to cite: FLORES, C. U. "Credit Card Fraud Detection Using Neural Networks". Codex Hash Research Lab, 2025. Available at: https://ulissesflores.com/research/2025-fraud-detection-mlp