Analyzing the effects of congestion on planning time index – Grey models vs. random forest regression

Abstract

Travel time and its predictability level are important factors influencing travel behavior and the role of traffic congestion as an important factor in the unreliability of travel time is undeniable. Traffic congestion is divided into two categories: Recurring and Non-recurring. In this paper, the effect of recurring congestion, defined as the ratio of the traffic speed over one hour to the free flow speed, will be investigated on the reliability of travel time in terms of the planning time index (PTI) on a 1.467-mile segment along the IS-64 freeway in Chesapeake, Virginia. To do so, two methods have been analyzed in this study: the grey models (GM) and the random forest regression (RFR). By using mean absolute percentage error (MAPE) as a criterion to judge, RFR could show a better and more satisfying performance in predicting PTI values when congestion changes. In the following, to make prediction results of RFR more understanding and easier to use, bagging and bootstrapping are used to improve the model results and more accurately predict the PTI. Then, the outputs were drawn using scatter plots for both peaks separately. Analyzing graphs has shown that for congestion values in the range of 1 to 0.9, PTI is reliable in both peaks. When congestion starts to decrease from 0.9 and reaches 0.7 or 0.75, depending on peak type, PTI is moving in the unreliable area, but it isn't extreme. Finally, when the congestion value becomes smaller, the rate of change in PTI in both peaks increases.