Using machine learning for bike sharing demand prediction

Abstract

Bike sharing has become a much more popular topic nowadays. Not only do the producers in bike-sharing need to provide a relatively accurate number of bikes in each period, but also the consumers need to have a general understanding of the number of bikes in each hour. This study analyses the dataset of bike-sharing rentals in 2011 in Washington, D.C. using machine learning, after training, testing, analyzing, and visualizing the dataset, the author chose the best model-random forest to predict it through the method of cross-test. The research result shows that the number of rentals in bike-sharing is the highest in the morning and evening travel peaks in one day, the highest in working days in one week, and the highest in autumn in one year. This information can help the bike-sharing service to prepare different quantities of bike-sharing at different times, and the customers would have a better overview of the bike demand when they plan to rent one. The whole research process provides valuable information for the service providers and users of bike-sharing.