Scholarlink

<p>Bike-sharing is a sustainable mode of transportation that can promote the transition to cycling. An essential part of bike-sharing management is to predict bike-sharing demand. This study aims to develop accurate models to predict bike-sharing outflow and inflow demand in Montreal, Canada. Although many factors can influence bike-sharing demand, only temporal and climate variables have been widely used to predict it. To address this, several other key variables are incorporated into the dataset, including land use, the built environment, cycling infrastructure, and socio-demographic variables. A powerful ensemble learning method, light gradient boosted machine (LightGBM), is used for modeling. Performance metrics indicate that the developed model performs better than models in the existing literature and that the added variables are crucial to the improvement. The other objective of this study is to capture the relation among the different variables and their impact on bike-sharing demand. Hence, two interpretation techniques are employed to better understand such influences. Shapley additive explanations and partial dependency plots are synchronized with LightGBM to identify the relative influence of variables on demand and capture the non-linear relationships between significant variables and bike-sharing demand. Findings suggest that cycling infrastructure at origins is more strongly associated with predicted demand than infrastructure at destinations. The highest demand is associated with neighborhoods with a higher walk score (over 95), a higher number of bike-sharing stations (over 3), a lower cycling distance to the center (within 5.5&#xa0;km), a higher temperature (over 15&#xa0;°C), and more and better cycling infrastructure (particularly at origins).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Towards Sustainable Urban Mobility: Interpretable Machine Learning for Bike-Sharing Inflow and Outflow Prediction

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Hamed Naseri,
  • Francesco Ciari,
  • Nicolas Saunier

摘要

Bike-sharing is a sustainable mode of transportation that can promote the transition to cycling. An essential part of bike-sharing management is to predict bike-sharing demand. This study aims to develop accurate models to predict bike-sharing outflow and inflow demand in Montreal, Canada. Although many factors can influence bike-sharing demand, only temporal and climate variables have been widely used to predict it. To address this, several other key variables are incorporated into the dataset, including land use, the built environment, cycling infrastructure, and socio-demographic variables. A powerful ensemble learning method, light gradient boosted machine (LightGBM), is used for modeling. Performance metrics indicate that the developed model performs better than models in the existing literature and that the added variables are crucial to the improvement. The other objective of this study is to capture the relation among the different variables and their impact on bike-sharing demand. Hence, two interpretation techniques are employed to better understand such influences. Shapley additive explanations and partial dependency plots are synchronized with LightGBM to identify the relative influence of variables on demand and capture the non-linear relationships between significant variables and bike-sharing demand. Findings suggest that cycling infrastructure at origins is more strongly associated with predicted demand than infrastructure at destinations. The highest demand is associated with neighborhoods with a higher walk score (over 95), a higher number of bike-sharing stations (over 3), a lower cycling distance to the center (within 5.5 km), a higher temperature (over 15 °C), and more and better cycling infrastructure (particularly at origins).