Search

Sensored traffic data in modern cities have been collected and applied for various purposes in the domain of intelligent transportation systems (ITS). However, analyzing these traffic data often lacks in priori knowledge due to the dynamics of transportation systems, making it hard to cope with diverse scenarios with specific models. In view of the limitations of traditional approaches, in this paper, we propose the Stepwise Heterogeneous Ensemble (SHE) for citywide traffic analysis based on stacked generalization. We first prove SHE's effectiveness using error-ambiguity decomposition technique. Secondly we analyze the optimal linear combination of SHE and present the stepwise iterating strategy. We also demonstrate its validity based on Kullback-Leibler divergence analysis. Thirdly we integrate six classical approaches into SHE framework, including linear least squares regression (LLSR), autoregressive moving average (ARMA), historical mean (HM), artificial neural network (ANN), radical basis function neural network (RBFNN), support vector machine (SVM). We further compare SHE's performance with other four linear combination models, namely equal weights method (EW), optimal weights method (OW), minimum error method (ME) and minimum variance method (MV). A series of experiments are conducted with a real city traffic dataset in Beijing city. The results show that the proposed SHE method behaves more robust and precise than other six single methods. Moreover, this method also outperforms other four different combination strategies both in variance and bias. In addition, the SHE method provides an open-ending framework for citywide traffic analysis, which means any new promising models can be easily incorporated into it in the future.