Keyword: Artificial Inteligence, Spatial Database, Spatial Modeling, Web Developement, Data Analysis
2022
Research
New York City Bike Share Demand Prediction explores how machine learning and spatial-temporal modeling can forecast hourly demand for shared bicycles at individual stations across Manhattan. The project develops a predictive framework to support more efficient redistribution strategies, helping balance bicycle availability and empty docks across the city’s network.
One of the persistent challenges in bike share systems is uneven bicycle distribution caused by one-way trips during peak commuting hours. This imbalance often leads to empty stations in high-demand areas and full docks in others, especially during morning and evening rush hours. Predicting demand accurately allows operators to relocate bicycles in real time and maintain reliable service throughout the city.
Using five weeks of Citi Bike trip data, the study built a space-time panel of over half a million records across 660 stations. Supplemented with weather, bike lane, and temporal data, five machine learning models were tested to predict hourly demand. The best model, combining spatial, temporal, and environmental features, showed strong accuracy and revealed clear rush-hour peaks and spatial clustering around central Manhattan’s bike lane network.
Conducted in 2022 as part of the Master of Urban Spatial Analytics (MUSA) program at the University of Pennsylvania, this project was completed under the guidance of Professors Elizabeth Delmelle, Michael Fichman, and Matt Harris.
The study successfully predicted hourly bike share demand with high accuracy despite being limited to a five-week dataset. Results highlighted clear spatial and temporal usage patterns, confirming peak demand during weekday commuting hours and reduced ridership in rainy or cold conditions. The research suggests that finer time intervals, such as 15-minute predictions, and real-time operational data from field managers could further enhance model accuracy.