Traffic forecasting is an important part of many future intelligent transportation systems and can be particularly useful for planning and navigation applications. The task is challenging because of the complex spatial patterns of road networks and dynamic temporal nature of traffic conditions. Recently, deep learning architectures with specially designed graph convolution layers to extract spatial patterns and recurrent or temporal convolution layers to extract temporal patterns have achieved good results for this task. In this paper, we propose a Mixture of Experts (MoE) based model integration framework to enhance the performance of these state-of-the-art traffic prediction models. In addition, we propose a novel entropy based loss function to improve the training of the MoE ensemble. Our experiments show that the performance of the Spatio-Temporal Graph Convolution Network (STGCN), a state of the art model, can be significantly improved.
Mixture of Experts based Model Integration for Traffic State Prediction
2022-06-01
745850 byte
Conference paper
Electronic Resource
English
Interpretable Cascading Mixture-of-Experts for Urban Traffic Congestion Prediction
ArXiv | 2024
|A Time Series is Worth Five Experts: Heterogeneous Mixture of Experts for Traffic Flow Prediction
ArXiv | 2024
|Traffic speed forecasting by mixture of experts
IEEE | 2011
|Competing risks mixture model for traffic incident duration prediction
Online Contents | 2015
|