An ensemble-learning method for potential traffic hotspots detection on heterogeneous spatio-temporal data in highway domain

• Published in: Journal of cloud computing : advances, systems and applications Vol. 9; no. 1; pp. 1 - 11
• Main Authors: Ding, Weilong, Xia, Yanqing, Wang, Zhe, Chen, Zhenyu, Gao, Xingyu
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 11-05-2020; Springer Nature B.V; SpringerOpen
• Subjects: Big Data; Computer Communication Networks; Computer Science; Computer System Implementation; Computer Systems Organization and Communication Networks; Domains; Edge computing security and privacy; Ensemble learning; Error reduction; Highway; Information Systems Applications (incl.Internet); Meteorology; Regression analysis; Security management; Software Engineering/Programming and Operating Systems; Spatio-temporal data; Spatiotemporal data; Special Purpose and Application-Based Systems; Traffic models; Traffic trends; Traffic volume; Trends; Big Data; Spatio-temporal data; Ensemble learning; Highway; Traffic trends
• ISSN: 2192-113X; 2192-113X
• DOI: 10.1186/s13677-020-00170-1

Inter-city highway plays an important role in modern urban life and generates sensory data with spatio-temporal characteristics. Its current situation and future trends are valuable for vehicles guidance and transportation security management. As a domain routine analysis, daily detection of traffic hotspots faces challenges in efficiency and precision, because huge data deteriorates processing latency and many correlative factors cannot be fully considered. In this paper, an ensemble-learning based method for potential traffic hotspots detection is proposed. Considering time, space, meteorology, and calendar conditions, daily traffic volume is modeled on heterogeneous data, and trends predictive error can be reduced through gradient boosting regression technology. Using real-world data from one Chinese provincial highway, extensive experiments and case studies show our methods with second-level executive latency with a distinct improvement in predictive precision.