Demand Analysis and Prediction of shared Bicycle Based on Machine Learning

As urban traffic continues to grow at a swift pace and road congestion intensifies, shared bicycles have increasingly caught the public's attention. Utilizing shared bicycles has emerged as a mode of transportation that is not only convenient and eco-friendly but also conducive to health. In this study, we employed two ensemble learning approaches—stacking and weighted averaging—to forecast the demand for urban bike-sharing services. These predictions were based on various optimized combinations of XGBoost, linear regression, and LightGBM models. Specifically, the stacking learning approach integrates models like XGBoost, linear regression, and LightGBM in diverse configurations. Meanwhile, the weighted averaging learning approach allocates distinct weights to models such as XGBoost, linear regression, and LightGBM. For the experimental phase, we utilized a bicycle rental dataset sourced from Washington D.C., USA, covering the period from 2011 to 2012. The evaluation metrics utilized were RMSLE, RMSE, and MAE. Our findings indicate that the ensemble model comprising XGBoost, linear regression, and LightGBM (referred to as XGB-LGB-LR) outperforms models employing alternative strategies across a range of metrics.