E-FLAPS: An Edge–Federated Learning Framework with Anomaly-Aware Training for Robust Stock Price Prediction
Abstract
Abstract Accurate and privacy-aware stock price prediction in distributed financial environments remains a challenging task, particularly under data heterogeneity and volatile market conditions. This paper presents E-FLAPS, a hybrid edge–Federated Learning (FL) framework that integrates BiLSTM networks with a Federated Anomaly-Aware Training (FAAT) strategy that embeds LOF-based anomaly filtering directly into each client’s local training loop, ensuring only reliable data contributes to federated model updates. The framework enables multiple financial institutions to collaboratively train a global model without sharing raw data, while edge-based preprocessing reduces latency and limits the propagation of noisy or anomalous patterns. Experiments conducted on realworld stock market data (2005–2024) across five major stocks (AAPL, MSFT, TSLA, NVDA, JPM) demonstrate that FAAT-based filtering alone reduces global MSE by approximately 11.54% over a standard FedAvg + BiLSTM baseline, and E-FLAPS achieves a 99.8% reduction in global MSE compared with the re-implemented M-A-BiLSTM baseline, driven predominantly by the federated architecture’s isolation of stock-specific dynamics — which prevents the cross-stock interference that causes M-A-BiLSTM to fail on NVDA — and further reinforced by FAAT-based anomaly suppression. Replacing FedAvg with FedProx reduces global MSE by a further 47.0%, confirming that proximal regularisation complements FAAT under non-IID financial data…
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.