Leveraging machine learning algorithms to estimate associated factors of students’ mode choice preferences for out-of-school-time activities

Abstract

Out-of-school time (OST) activities are essential for students’ educational achievement and workforce development. However, economically disadvantaged youth have far less OST access compared to their wealthier counterparts. In sprawling, low-density metropolitan areas, a lack of reliable and cost-effective transportation services constitutes a considerable barrier to OST participation for youth from these families. Research on youth travel focuses on school trips and compares active travel modes, such as biking and walking, to motorized modes. However, there is a gap in understanding the travel choices, preferences, and perceptions of underrepresented youth to access OST activities. This study conducted a revealed preference and stated preference survey of underrepresented youth in metropolitan Kansas City, KS-MO, aged 13–18 years old, to capture their travel behavior and interest in utilizing different modal options and evolving technologies in transportation to access OST activities. Five realistic Stated Choice Experiments pivoting the level of travel time and cost for each alternative (i.e., sharing a ride, hailing a ride, renting bikes/scooters, and taking public transit) were conducted. The study used several machine learning (ML) algorithms to predict students’ mode choice preferences at aggregated and individual levels. Among the ML algorithm approaches deployed here, the boosting-based ensemble learning models outperformed the other ML models. The results showed that most students had a low probability of switching from their current mode of transportation across the different choice scenarios, possibly due to a lack of equitable and sustainable transportation options for students to engage in OST activities and uncertainty in the role of evolving technologies in transportation to address this lacking.