Optimizing Urban Traffic Congestion Through Artificial Intelligence and Machine Learning: A Case Study of Dar Es Salaam City

Abstract

Urban traffic congestion remains a defining obstacle to efficient mobility and infrastructure use in rapidly growing cities. In Dar es Salaam, particularly along Morogoro Road, static lane configurations are poorly matched to dynamic, directional traffic patterns, producing peak-direction bottlenecks and underutilised capacity in the counterflow. This study evaluates a Dynamic Lane Allocation Logic (DLAR) coupled with a Utilisation Efficiency (UE) metric to optimise lane use in real time. Using observed directional volumes and a standardised urban lane capacity of 1,800 pcu/h, we operationalise UE to compare baseline static (2+2), dynamic (e.g., 4+1), and AI-adaptive configurations (up to 5+1). Scenarios are implemented and tested in the Simulation of Urban Mobility (SUMO) platform to ensure reproducible, controlled experiments. Results indicate that dynamic configurations substantially improve efficiency, with AI-adaptive allocations achieving UE values of up to 95% during peak periods, compared to approximately 60% under static control. These findings underscore the value of data-driven, context-aware lane management and support near-term pilots of AI-based traffic control, investment in real-time sensing and monitoring infrastructure, and the adoption of localised capacity benchmarks in planning practice. For sub-Saharan African cities facing similar constraints, the approach offers a pragmatic pathway to relieve congestion without costly physical expansion, aligning day-to-day operations with the realities of fluctuating urban demand.