Traffic Forecasting Using Federated Randomized High-Order Fuzzy Cognitive Maps
Resumo
Numerous machine learning (ML) and deep learning (DL) forecasting techniques have emerged for predicting traffic flow and traffic speed. Despite their notable achievements, they encounter challenges related to preserving in-vehicle user privacy. Alternatively, Federated Learning (FL) offers a decentralized ML strategy to strike a balance between prediction accuracy and privacy preservation. This paper proposes a novel fuzzy-based method called FL-RHFCM, which integrates the principles of Randomized High-order Fuzzy Cognitive Maps (R-HFCM) with FL. R-HFCM is akin to an echo state network (ESN), where the only trainable component is the output layer using least squares (LS) minimization. The reservoir layer’s weights are initialized randomly and remain unchanged during training. Thus, the LS coefficients represent the only parameters shared with the server for aggregation in our FL-RHFCM approach. FL-RHFCM introduces an efficient distributed forecasting method using FCMs, significantly reducing communication costs. The effectiveness of our proposed model is assessed using two datasets, demonstrating its promise compared to some existing baseline methods.
