Leveraging machine learning to predict Rural-Urban carpooling decisions given monetary incentives

Abstract

This study pioneers a quantitative method for determining the monetary incentives needed to promote carpooling in an area. Utilizing travel behavior survey data from a rural county in Central Pennsylvania, we employed a Random Forest machine learning model to predict individuals’ carpooling decisions under different incentive scenarios based on their sociodemographic characteristics and existing travel behaviors. We then calculated the pivotal incentive amount that encourages an individual to switch from solo driving to carpooling and estimated the total incentive budget required to boost carpooling participation in an area to a certain desired level. Our analysis revealed that high spending on incentives does not necessarily result in proportional increases in carpooling participation. For instance, for our study area, the effective monthly incentive amount was capped at approximately $300. We also found that there are individuals who will not carpool at all, regardless of the incentive amount offered. Additionally, by clustering respondents based on their sociodemographic and commuting characteristics, we investigated how different groups of population respond to incentives. The results showed that younger adults, solo workers in a household, and those who commute longer and more frequently required the highest incentive amounts to switch to carpooling, whereas seniors and those with less frequent or shorter commutes could be swayed with incentive amounts about half as much. Such findings highlighted the importance of targeted strategies with detailed behavioral analysis when determining and allocating incentive funds. Finally, high performance metric values for the model demonstrated the potential of machine learning as a viable forecasting tool for similar transportation studies.