Optimizing High-Frequency Risk Control Decision-Making Mechanisms in FinTech Scenarios Using the FEDformer Model
J. Y. Huo · Advanced Electromagnetics · 2026
Efficient risk identification in high-frequency financial systems requires simultaneous optimization of inference latency and predictive accuracy, posing challenges similar to those encountered in real-time signal processing and frequencydomain analysis for communication systems. This study proposes a scenario-oriented high-frequency risk control framework based on the Frequency Enhanced Decomposed Transformer (FEDformer) to improve decision-making performance in FinTech environments. By integrating an adaptive frequency decomposition module, a lightweight frequency-enhanced attention mechanism, and an incremental training strategy, the proposed framework effectively suppresses multi-scale noise while preserving critical temporal patterns in high-frequency transaction streams. Furthermore, a multi-scale feature fusion architecture and hybrid loss function are introduced to address class imbalance and enhance risk representation under dynamic operating conditions. Experimental evaluation demonstrates that the optimized model achieves an AUC of 0.942 with an inference latency of 12 ms, outperforming conventional Transformer-based approaches in both detection accuracy and computational efficiency. Deployment in practical payment scenarios further improves fraud interception rates while reducing false rejection rates and operational costs. Beyond financial risk management, the proposed frequency-aware modeling strategy provides methodological insights for intelligent electromagnetic signal analysis, adaptive spectrum feature extraction, and low-latency decision systems in communication-oriented applications where robust frequency-domain representation and real-time processing are essential.