Portal de Periódicos da CAPES

CO-STAR: A collaborative prediction service for short-term trends on continuous spatio-temporal data

2019; Elsevier BV; Volume: 102; Linguagem: Inglês

10.1016/j.future.2019.08.026

1872-7115

Weilong Ding, Xuefei Wang, Zhuofeng Zhao,

Data Stream Mining Techniques

Over various sensory data of Internet of Things, not only the current situation but also the future trends of many fields are required instantly to promote the business. As a typical requirement, the short-term prediction on spatio-temporal data stream is imperative, but challenges still remain due to the inherent limitation of long calculative time and insufficient predictive precision. In this paper, a novel prediction service CO-STAR is proposed in the highway domain. On the continuous toll data of the whole highway network, the service employs non-parametric regression model to predict the traffic volume of all the stations periodically. Considering both spatial and temporal business characteristics, a collaborative paradigm of online stream computing and offline batch processing is adopted to balance the efficiency and precision. On the real data of one Chinese provincial highway and the simulated data, our service can hold minute-level executive latency with nearly 10 percent improvement for the predictive precision in extensive experiments.