Divvy Bike-Share.zip

This study builds a simple and accurate forecasting model to predict monthly <b>Divvy bike-share trip volumes</b> and <b>average trip durations</b> in Chicago. Using data from January to December 2024—including trip records, weather conditions, and neighborhood characteristics—the researchers created nine features (like weekend flags and past usage data) for each community area.They trained a <b>LightGBM model</b> in R and tested it using two methods to avoid bias and ensure it works well across different neighborhoods. The final model predicted trip counts with an error of only <b>8.4%</b> and trip durations with <b>2% error</b>. Most of the prediction power came from the <b>one-month lag</b> in usage.The model proves that a <b>simple, interpretable approach</b> (not deep learning) can give <b>urban planners accurate insights</b> for bike-share demand forecasting.

本研究构建了一款简洁且精准的预测模型，用于预测美国芝加哥市Divvy共享单车（Divvy bike-share）的月度出行量与平均出行时长。研究人员采用2024年1月至12月的多源数据集，涵盖出行记录、气象条件与社区街区特征，为每个社区区域生成了9项特征（例如周末标识、过往使用数据）。研究者在R编程语言环境中训练了LightGBM模型，并通过两种测试方法规避模型偏差，确保其在不同社区区域均具备可靠的预测性能。最终模型的出行量预测误差仅为8.4%，出行时长预测误差为2%。其中，出行量的单月滞后（one-month lag）特征贡献了绝大多数的预测效能。本研究证实，无需依赖深度学习，采用简洁可解释的建模思路，即可为城市规划者（urban planners）提供精准的共享单车需求预测洞察。