Assessing workplace relocation impacts on travel time reliability: an activity-based modeling framework with machine learning-based trip duration prediction

Rapid and often unplanned urban growth in African secondary cities is placing increasing pressure on transport systems, resulting not only in congestion but also in unstable and unreliable travel times. In cities such as Bahir Dar, Ethiopia, strong spatial concentration of employment and services produces directional travel demand and uneven network performance. However, African urban transport studies rarely combine behaviorally detailed simulation with data-driven predictive analysis to evaluate land-use restructuring strategies. This study combines Activity-Based Modeling (ABM) and Machine Learning (ML) within a unified scenario analysis framework. The ABM is used as the primary scenario engine to simulate behavioral responses under alternative workplace distributions and assess the effects of workplace relocation and land-use redistribution on travel patterns and travel time reliability, while ML models are applied to predict trip durations from ABM-generated trip records using departure time and transport mode as predictor variables. Scenario results show that while employment decentralization improves the spatial balance of trips, it can increase travel time variability and upper-tail travel time risk due to network load redistribution effects, indicating measurable polycentric trade-offs rather than automatic reliability gains. The ML analysis further showed that ensemble tree-based models outperformed the MLP model in trip-duration prediction, with Random Forest achieving the highest predictive performance across the evaluated scenarios. Methodologically, the study helps address a gap in African urban transport modeling by linking activity-based behavioral simulation with ML-based trip-duration prediction within a unified scenario analysis framework. Practically, the results provide quantitative evidence to support Ethiopian urban planners and policymakers in evaluating decentralization and sub-center development strategies under resource and infrastructure constraints. The framework may be applicable to other rapidly urbanizing African cities, subject to the availability of local data and appropriate model calibration and adaptation.