Station-level Demand Prediction for Bike-Sharing System

Bike-sharing programs have received increasing attention in the recent years due to their positive impact on the environment coupled with the awareness created by the environmental activists. In addition to the environmental advantages, cycling aids in improving public health, increasing cycling population and contributing to transit use. With the increasing number of bike-sharing service providers around the world, a real-time demand-supply tracking system which brings together the service providers and bike enthusiasts is implemented. Though there are several bike-sharing applications readily available for the users, currently there are no apps for the bike-sharing owners to track the demand-supply for bikes in real-time for any given station. We implemented a real-time demand-supply system that provides to the bike owner information regarding the demand-supply for bikes in any particular station at any particular time through a dashboard and also does forecasting of the bike demand. The Bike Demand forecasting uses the following classifiers: random forest, gradient boosting and linear regression, to predict the number of bikes in demand at a particular station in real-time. Results from the above-mentioned methods are compared and the method with the best results is chosen to supplement information provided by the dashboard to show the demand. For all three of our datasets, XGBoost on the dataset with all stations outperformed all the other models and hence XGBoost on trips data was chosen for the web application. Bike share providers will know the demand for any particular station which will enable them to fetch bikes from stations using the Web UI. Bike shortages due to uneven bike distribution are avoided with high accuracy due to fast prediction.