Broad Learning for Optimal Short-Term Traffic Flow Prediction

Abstract

In this work, we explore the use of a Broad Learning System (BLS) as a way to replace deep learning architectures for traffic flow prediction. BLS is shown to not only outperforms standard learning algorithms (Least absolute shrinkage and selection operator (LASSO), shallow and deep neural networks, stacked autoencoders) in terms of training time, but also in terms of testing accuracy.