Demand Prediction on Bike Sharing Data Using Regression Analysis Approach

Abstract

In order to forecast the need for bike-sharing services, this paper suggests a rule-based regression model. Commuters and tourists alike are taking advantage of public bike sharing programs because of the convenience and low carbon footprint they provide. Used information from the UCI Machine Learning Repository. Repeated cross-validation was used to fine-tune the hyper-parameters of five statistical models. Conditional Inference Tree, K-Nearest Neighbor Analysis, Regularized Random Forest, Classification and Regression Trees, and CUBIST. The predictive accuracy of the regression models was measured by calculating the Root Mean Squared Error, R-Squared, Mean Absolute Error, and Coefficient. For both the Seoul Bike and Capital Bikeshare programs, the rule-based model CUBIST was able to account for 95 and 89% of the Variance (R2), respectively. All models built from the two datasets using WEKA v3.8.6, and are used a variable significance analysis to establish which variables were most crucial. The most important factors in determining the hourly demand for bike rentals are the weather and the time of day.