Predictive Modeling of Passenger Tipping Behavior for Yellow Taxis and For-Hire Vehicles in New York City Using Machine Learning Techniques

This study investigates predictive methodologies for passenger tipping behavior associated with yellow taxis and For-Hire services in New York City. By evaluating four machine learning models—Linear Regression, Ridge Regression, Decision Tree, and Random Forest—the research examines and assesses the predictive accuracy of different models concerning passenger tip amounts. The findings indicate that the parameter-tuned Random Forest model exhibited superior performance across both the yellow taxi and FHV datasets, evidenced by the lowest mean square error (MSE) on the test set, achieving values of 1.4286 and 8.2790, respectively. This demonstrates the model's exceptional predictive accuracy and generalization capabilities. The contribution scores of each feature in the prediction of the model were calculated, and it is found that the two variables of trip distance and congestion surcharge are the factors that contribute most to the prediction of passenger tipping behavior. In addition, the regression coefficient analysis also shows that the trip distance and trip duration have a positive impact on the tipping payment of both modes of transportation, while the weather conditions have a significant difference in the impact of passenger tipping on For-Hire Vehicle services and yellow taxis. This study provides valuable empirical evidence for the optimization of urban transportation operations, helps urban planning managers and taxi and online ride-hailing operators to deepen their understanding of tipping behavior, and thus provides a reference for the optimization of future transportation efficiency and policy formulation.