Factors affecting rider churn in on-demand food delivery services: insights using a survival analysis and interpretable machine learning approach
研究利用生存分析和可解释机器学习，分析某外卖平台数据，找出影响骑手流失的关键运营因素，为平台管理者提供改善骑手留存的建议。
The rapid expansion of on-demand food delivery (ODFD) platforms has intensified concerns about the sustainability of gig work, particularly due to high rider churn. This study investigates operational-level drivers of rider churn in the ODFD sector using a survival analysis framework enhanced with interpretable machine learning (IML) techniques. A right-censored dataset from a leading ODFD platform comprising delivery records, weather, traffic conditions, and rider activity is analysed. Traditional survival models are limited by strict assumptions that are difficult to verify in partially concealed datasets. To overcome this, advanced machine learning-based survival models are utilised and compared using the concordance index to select the optimal algorithm. To improve transparency in machine learning algorithm outputs, IML tools, such as feature importance and partial dependence plots, are used to identify key factors influencing rider attrition. The findings highlight critical operational factors driving churn, offering actionable insights for platform managers seeking to improve rider retention and reduce inefficiencies. This study contributes methodologically by integrating machine learning with survival analysis, in combination with IML tools. It also advances empirical understanding of micro-level dynamics affecting gig worker sustainability. The approach offers a robust decision support framework to address workforce instability in platform-based service ecosystems.