Short-Term Traffic Flow Forecast Using Regression Analysis and Graph Convolutional Neural Networks

Short-term forecast of different traffic attributes is one of the fundamental tools used in transportation planning. Precisely, accurate predictions of traffic flow are often the basis of an efficient traffic management. In the paper at hands we shed a light on short-term traffic flow forecast in urban arterial roads in Bremen (Germany). This case-study uses real-data collected from 7 loop detectors installed in downtown. To deal with this, we propose two different models, namely, Linear Regression and Graph Convolutional Neural Networks. The models are separately applied to 11 weeks of data, three of these are dedicated to test the performance of both models. Experimental results show that the models are closely competitive and they reach, in average over the whole test-set, around 19% mean absolute percentage error during morning and evening peak times.