A Hybrid Machine Learning Model for Demand Prediction of Edge-Computing-Based Bike-Sharing System Using Internet of Things

Abstract

The rapid development of Internet-of-Things technologies (such as edge computing) has promoted the development of numerous emerging urban applications, particularly smart transportation. As an anticipated aspect of smart transportation, bike-sharing systems have recently been deployed in many cities and are considered an efficient way to address the issue of “the last mile.” In a bike-sharing system, the supply and demand of shared bikes at each bike station frequently change over time. Consequently, one of the most challenging issues of a bike-sharing system is predicting the required number of shared bikes at each station. In this article, we take the real aspects of a bike-sharing system into account, e.g., the high complexity, nonlinearity, and uncertainty of the traffic flow, and propose a hybrid edge-computing-based machine learning model. Notably, our proposed model, which combines a self-organizing mapping network with a regression tree (RT), is applied to predict the bicycle demand of a certain station through the following steps: 1) the proposed model adopts self-organization mapping to assemble the original samples in the form of clusters and 2) each cluster is then built as an RT to forecast the required number of bikes at each station. Experiments based on real data from the Washington and London bike-sharing systems show that our proposed method achieves a higher prediction accuracy and better generalization than previous approaches.