Adaptive Privacy-Budget Federated Ensemble with Meta-Learning and Contrastive Learning for Rare Fraud Detection
Abstract
In multi-institutional financial risk control, detecting rare fraudulent behavior under strict privacy rules is difficult because of data imbalance, limited sharing, and changing patterns over time. This paper presents FedPrivBoost, a privacy-enhanced federated ensemble framework that uses differential privacy, meta-learning, and contrastive representation learning to improve model stability and adaptability under privacy limits. The framework applies a private meta-learning method to adapt across different institutions, a heterogeneous ensemble trained under Rényi differential privacy to improve generalization, and a contrastive module with privacy-based data augmentation to learn clear features for rare fraud samples. It also includes an adaptive privacy-budget allocation based on Fisher information and Bayesian optimization to balance model performance and privacy during training. FedPrivBoost gives a simple and reliable solution for secure and effective fraud detection in federated environments.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.