Work Zone Capacity Estimation Using an Ensemble Tree Approach
Weng, Jinxian; Meng, Qiang
Accurate estimation of work zone capacity is critical to the success of traffic control management strategies at work zones. This study uses an ensemble tree approach to more accurately estimate work zone capacity compared to the existing single decision tree based model. A bootstrap aggregation method is employed to build an ensemble tree comprising a set of individual decision trees. More specifically, a set of bootstrap samples are first generated by sampling with replacement from a training sample. With these bootstrap samples, a set of individual trees are then constructed by using a tree learning algorithm and combined by averaging the output. Based on the work zone capacity data from 14 states and cities, one case study is finally conducted to build and evaluate the ensemble tree in this study. The statistical comparison results fully demonstrate that the ensemble tree outperforms the existing work zone capacity models in terms of estimation accuracy. The ensemble tree also performs better than any of single decision trees in terms of stability. The comparison with the new version of Highway Capacity Manual (HCM2010) indicates that the ensemble tree can provide a more accurate estimate of work zone capacity. Unlike the HCM, the ensemble tree avoids estimated errors caused by users’ subjective judgments because it does not require manually setting various adjustment factors. Because of the high estimation accuracy and stability, the ensemble tree is a good alternative to estimate work zone capacity, especially for those inexperienced users.
Presented at the 91st Annual Meeting of the Transportation Research Board, January 2012, Washington, D.C.