A Comparative Study of Machine Learning Models for Bike Sharing Demand Prediction: A Feature-Centric Approach

Abstract

As smart city initiatives advance globally, precisely forecasting bike sharing demand has emerged as a critical challenge for optimizing urban transit systems and allocating resources efficiently. This study aims to compare the performance of three classical machine learning models — Linear Regression, Decision Tree, and Random Forest — in predicting hourly bike rental counts based on time and environmental features. Leveraging the publicly available Kaggle Bike Sharing Demand dataset, this study carries out thorough experiments incorporating systematic feature construction and temporal cross-validation. The experimental results show that the Random Forest model achieves the best overall performance with RMSE of 14.44, MAE of 10.73, and R² of 0.8922, outperforming Decision Tree (RMSE: 17.40, R²: 0.8433) and Linear Regression (RMSE: 21.03, R²: 0.7711). Feature importance analysis reveals that hour-related cyclic features (hour_sin) are the most influential factor, with an importance score of 0.602, followed by hour (0.172) and temperature (0.073). This study provides empirical evidence for model selection in bike sharing prediction tasks and offers practical guidance for urban transportation management.