Abstract
This study proposes a tensor-based K-Nearest Neighbors (K-NN) method, in which traffic patterns involve multi-dimensional temporal information and bi-directional spatial information. Such multi-temporal information can not only capture the instantaneous fluctuation of short-term traffic but keep the general trend of long-term traffic. In numerical experiments, with taxis’ GPS data from an urban road network, traffic speed data are organized into one- (2 min), two- (4 min) and three- (2, 4 and 10 min) temporal dimensions. Meanwhile, spatial information about six upstream links and six downstream links of the target link is incorporated to construct the tensor-based data structure. Numerical results show that the K-NN with three temporal dimensions (K-NN 3D) outperforms other methods under no data missing or under various random/module/mixed data missing rates. In summary, the tensor-based K-NN method is promising in the traffic prediction under data missing cases.
Disclosure statement
No potential conflict of interest was reported by the authors).