Customer booking prediction is essential for On-Demand Transport services, especially for those in rural and suburban areas where the demand is low, variable and often regarded as unpredictable. Existing literature tends to focus more on the prediction of demand for traffic, classical public transport, and urban On-Demand Transport service such as taxi, Uber or Lyft, in areas with higher and less variable demand, in which popular time-series prediction methods can be employed. This paper proposes an ensemble learning framework to predict the customer booking behaviour and demand using the observed data of a suburban On-Demand Transport service where data scarcity is a challenge. The proposed method, which is called as Class-specific Soft Voting, is found to be the most accurate prediction method when compared to popular supervised classification methods such as Logistic Regression, Random Forest, Support Vector Machine and other ensemble techniques.

客户预订预测对于按需运输服务至关重要，尤其是对于需求低，变化多端且经常被认为不可预测的农村和郊区地区的需求。现有文献倾向于在需求变化较大和较小的区域中，将重点更多地放在交通需求，经典公共交通和城市按需交通服务（如出租车，Uber或Lyft）的需求预测上，其中流行的时间序列预测方法可以雇用。本文提出了一个集成的学习框架，可以使用郊区按需运输服务的观测数据来预测客户预订行为和需求，而数据稀缺是一个挑战。所提出的方法称为类特定的软投票，