Inferring travel time preferences through a contextual feature fusion approach

Digital navigation services are extensively employed to provide travelers with recommendations for reaching their destinations. However, most current navigation services primarily focus on time and distance when suggesting routes, neglecting the consideration of the value of travel time (VTT). VTT represents a mobility paradigm that recognizes travel time as an opportunity for various activities, such as work tasks or leisurely pursuits like listening to music. The incorporation of VTT facilitates the provision of personalized recommendations tailored to travelers’ individual preferences. In this study, we assess travelers’ VTT using four distinct elements: paid work, personal tasks, enjoyment, and fitness. To infer VTT, we propose an innovative approach that fuses features extracted from different contexts, including physical conditions (e.g., weather) and traveler attributes (e.g., gender, age). These extracted features are then input into our suggested machine learning framework, which comprises boosted decision trees and deep learning Transformers. The results demonstrate that our framework provides the most accurate VTT predictions when compared to traditional machine learning models and rule-based baselines. Additionally, the analysis of travelers’ VTT predictions reveals several intriguing patterns that contribute to a better understanding of their decision-making process when selecting a travel route.