Scholarlink

Mobile payment systems like Bankily, Masrvi, and SEDAD have expanded financial inclusion in Mauritania to 38% by 2023. However, rising cybercrime—with 1,276 complaints and 501 million MRU ($12.65M USD) in losses during Q1 2025—threatens user trust. We propose a pioneering multi-layered AI framework integrating real-time fraud detection, facial and fingerprint biometric authentication, and intrusion detection, specifically tailored to Mauritania’s FinTech ecosystem. Our models, trained on PaySim, IEEE-CIS, NSL-KDD, and CICIDS2017 datasets, achieve state-of-the-art results: fraud detection AUC-ROC 0.97 ± 0.01, precision 0.99, recall 0.81, biometric EER 0.02 (FaceNet) and 0.03 (ResNet18), and IDS F1-scores up to 0.95 (DoS) and 0.92 (SQL Injection). Compared with traditional baselines, our framework reduces false positives by 30% while maintaining high recall. Ethical safeguards include AES-256 encryption and GDPR-compliant consent. Deployed via edge computing, the system offers 2–5 ms response latency and scales across African FinTech infrastructures. The project was awarded First Place in the AI Bridge Challenge 2025 (GIZ – I2COMSAPP 25) for its real-world entrepreneurial impact, providing a robust blueprint for secure, inclusive FinTech ecosystems in developing regions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI for Secure Mobile Payments: A Multi-layered Framework for Mauritania

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Cheikh Abdelkader Ahmed Telmoud

摘要

Mobile payment systems like Bankily, Masrvi, and SEDAD have expanded financial inclusion in Mauritania to 38% by 2023. However, rising cybercrime—with 1,276 complaints and 501 million MRU ($12.65M USD) in losses during Q1 2025—threatens user trust. We propose a pioneering multi-layered AI framework integrating real-time fraud detection, facial and fingerprint biometric authentication, and intrusion detection, specifically tailored to Mauritania’s FinTech ecosystem. Our models, trained on PaySim, IEEE-CIS, NSL-KDD, and CICIDS2017 datasets, achieve state-of-the-art results: fraud detection AUC-ROC 0.97 ± 0.01, precision 0.99, recall 0.81, biometric EER 0.02 (FaceNet) and 0.03 (ResNet18), and IDS F1-scores up to 0.95 (DoS) and 0.92 (SQL Injection). Compared with traditional baselines, our framework reduces false positives by 30% while maintaining high recall. Ethical safeguards include AES-256 encryption and GDPR-compliant consent. Deployed via edge computing, the system offers 2–5 ms response latency and scales across African FinTech infrastructures. The project was awarded First Place in the AI Bridge Challenge 2025 (GIZ – I2COMSAPP 25) for its real-world entrepreneurial impact, providing a robust blueprint for secure, inclusive FinTech ecosystems in developing regions.