Demand forecasting for bike sharing based on Bayesian XGBoost

Abstract

A prediction model of shared bicycle rental volume based on the Bayesian-XGBoost algorithm is developed to address the issue of the imbalance between the supply and demand of shared bicycles. Correlation analysis is first used to investigate the characterization, followed by hyperparameter tweaking and assessing the model's stability and adaptability. The comparison study confirms that the Bayesian-optimized model has a greater forecast accuracy than a single model in shared bicycle rental volume prediction. The tests employ shared bicycle trip data in Washington, D.C., U.S.A.