Short-Term Bus Passenger Flow Forecast Based on the Multi-feature Gradient Boosting Decision Tree

Accurate prediction of bus passenger flow is an important basis for the dynamic scheduling of urban smart transportation system. In order to accurately predict short-term bus passenger flow and help managers achieve efficient dispatching operations and alleviate traffic pressure, a Multi-feature Gradient Boosting Decision Tree (GBDT) model is proposed. Using the flexibility of Gradient Boosting Decision Tree algorithm in complex data processing, a Gradient Boosting Decision Tree basic model is established. Through data analysis and processing, multiple features such as the week, time and environmental factors that related to passenger flow were mined to construct effective feature engineering, which can make the model’s prediction results more accurate. Experiments running on the real data set of Guangzhou show that the multi-feature Gradient Boosting Decision Tree model can predict bus passenger flow accurately and effectively. The MAPE of the model is 1.152%, the RMSE is 4.58, under the same conditions, which are better than the prediction effect of traditionally used models such as the Linear Regression and the BP Neural Network, etc.