A Data Driven Approach to Forecasting Traffic Speed Classes Using Extreme Gradient Boosting Algorithm and Graph Theory

Historical cities around the world have serious traffic congestions due to old infrastructure and urbanization. To mitigate traffic problems in such cities, infrastructure investments are channeled to tunnels, bridges, and highways. The solution becomes more complicated as the city centers are expected to become fully pedestrian-friendly as a European Union target for 2050. Any modification to the city transport network will lead to changes in traffic density patterns. Investment decisions become more pronounced as the speed of urbanization increases. However, this decision-making processes in urban transport networks pose a serious risk to city managers as wrong decisions are expensive and will not solve the problem. This paper proposes a practical, cost-effective model to assist decision makers by providing them with estimated changes in the traffic patterns due to the addition of new roads to existing infrastructure. The study adopts the Extreme Gradient Boosting (XGboost) which is a tree-based algorithm that provides 85% accuracy for estimating the traffic patterns in Istanbul, the city with the highest traffic volume in the world. The proposed model is a static model that allows city managers to perform efficient analyses between projects that involves changes to the city's transport network.