Public bike-sharing (PBS) systems have expanded to major cities around theworld in efforts to mitigate air pollution, traffic congestion and trafficaccidents. Users can pickup and drop-off bicycles at any station, and thusinventory imbalances can occur. To improve system efficiency, systemoperators should establish appropriate repositioning strategies based onaccurate predictions of demand for bicycles. This study aims to predictstation-level demand for pickup and drop-off of bicycles using stationactivity information. In addition to time and weather information, thenumber of pickups and drop-offs at a station 1–3 h before the predictionwas used as a predictor. A random forest machine learning technique isadopted for the demand prediction. The PBS database in Seoul, SouthKorea was used for the case study. To compare prediction accuracy bystation usage patterns, the stations are classified into four clusters. Theanalysis results show that prediction accuracy including lag informationprovides mprovements of up to 20%, and the forecast for drop-off is moreaccurate than the forecast for pickup. This study practically contributes toincreasing operational efficiency and reducing operating costs by improvingdemand predictability in a PBS system.
展开▼