Deep Learning for Short-Term Prediction of Available Bikes on Bike-Sharing Stations

Bike-sharing is adopted as a valid alternative to traditional public transports since they are eco-friendly, prevent traffic congestions, reduce the probability of social contacts which happens on most means. However, some problems may occur such irregular distribution bikes stations/racks/areas, and difficulty knowing in advance rack status with certain degree confidence, whether there will be available at specific bike-station time day, or free slot for leaving rented bike. Thus, providing predictions can useful improve quality service, especially those cases bike racks used e-bikes, need recharged. This paper compares state-of-the-art techniques predict number bike-slots bike-sharing stations (i.e., racks). To this end, set features predictive models were compared identify best predictors short-term 15, 30, 45, 60 minutes. The study demonstrated that deep learning particular Bidirectional Long Short-Term Memory networks (Bi-LSTM) offer robust approach implementation reliable fast bikes, even limited amount historical data. also reported an analysis feature relevance based SHAP validity model different behaviours clusters. solution its validation derived by using data collected bike-stations cities Siena Pisa (Italy), context Sii-Mobility National Research Project Mobility Transport Snap4City Smart City IoT infrastructure.