Modelling the Influence of Departure Time on Trip Duration and Mode Choice: An Activity-Based Model and Machine Learning Approach

Bahir Dar City, Ethiopia, is experiencing rapid urbanization and motorization, leading to major transportation challenges characterized by uneven travel demand and variable trip durations. This study investigates how departure times influence trip duration and mode choice by integrating activity-based models (ABMs) with machine learning techniques. The combined approach enhances predictive accuracy and offers deeper insights into the temporal dynamics of travel behavior. Using population synthesis with PopulationSim and activity execution modelling in VISUM, a total of 147,351 activities, 56,121 tours, and 112,429 trips were generated. Results reveal strong temporal variations in mobility, with peak demand observed around 8 AM, 12 PM, 1 PM, 2 PM, and 5 PM. Minibuses dominate urban mobility during these periods, while auto-rickshaws provide flexible options for medium-distance travel. Analysis of trip durations shows that buses and minibuses experience the greatest variability due to congestion and operational delays, whereas private cars, walking, and auto-rickshaws demonstrate greater stability. Machine learning results indicate that Random Forest (R² = 0.769, RMSE = 3.78) outperforms XGBoost and MLP in predicting trip durations and capturing peak-hour dynamics. These findings underscore the critical role of departure time in shaping travel outcomes and suggest the need for coordinated land use and transport policies to balance demand, improve reliability, and support sustainable mobility.