大数据驱动下的共享单车短期需求预测——基于机器学习模型的比较分析

Short Term Bike-sharing Ridership Prediction under the Big-data Condition:Comparison of Machine Learning Models

基于共享单车项目的多维度大样本数据,以套索回归、岭回归、随机森林和迭代决策树等机器学习模型,探讨了共享单车短期(基于小时)需求预测的主要影响因素,并对不同模型预测效果进行了比较分析。研究结果发现,影响共享单车小时需求的主要因素包括特定的位置因素、时间因素以及天气条件因素。同时,相比普通线性回归、套索回归和岭回归模型,随机森林和迭代决策树模型对共享单车短期即时需求预测的结果更精确,在样本内部拟合和样本外推预测中的拟合优度(R2)更高,标准误差(RMSE)更低,是共享单车行业短期实时需求精准预测的更有效手段。

Using the large and multidimensional data released by the bike-sharing project,and employing the Machine Learning Models,this article discussed the factors influenced short-term demand prediction of a bike-haring business.The results showed that the major factors that affected the short-term demand of bike sharing include specific location,time,and weather conditions.Meanwhile,compared with the Ordinary Linear Regression,Lasso Regression and Ridge Regression model,the Random Forest and Gradient Boosting Decision Tree models had higher goodness of fit(R 2)and lower standard error(RMSE)in both in sample and the out sample predictions,which shed lights on the machine learning models and are more suitable for short-term precise demand predictions.