Short-term demand forecasting for bike sharing system based on machine learning

This paper takes the New York Citi Bike public bicycle system as the research object, and uses the public bicycle historical operation data, weather data, and holiday information to propose a short-term demand forecasting method for the total lending amount of the stations in the next hour based on machine learning approaches. This study extends the existing literature in the following three aspects: (1) we clustered bike stations creatively by a new spatial constrained clustering method; (2) it used the nested sliding window method to reduce the over-fitting risk of the prediction model and verifies the accuracy of the model prediction results; (3) we exacted new time series features and applied them to the linear prediction model, the prediction results are compared with the nonlinear models, and it is found that the linear model outperformed the non-linear model in predicting station demand.