Investigating the Impact of Weather on Demand Prediction for Bike Sharing System

Rebalancing a bike sharing system involves removing bikes from oversubscribed stations and putting them into undersubscribed stations in order to satisfy demand. Predicting the number of spaces required depending on day, time and weather is a challenge. This research proposes to investigate the two most prevalent prediction methods in conjunction with weather data to find whether clustering or tree methods provide a better model for improving demand prediction to aid system rebalancing. The previous research on DC Bikes was replicated, in Dublin however weather has much less of an impact on Dublin Bikes usage. Random Forest Classification gave better demand prediction for rebalancing the bike system at a station level and a model that combines the two methods may well be better overall than either individually