TY - GEN
T1 - Spatiotemporal periodical pattern mining in traffic data
AU - Jindal, Tanvi
AU - Giridhar, Prasanna
AU - Tang, Lu An
AU - Li, Jun
AU - Han, Jiawei
PY - 2013
Y1 - 2013
N2 - The widespread use of road sensors has generated huge amount of traffic data, which can be mined and put to various different uses. Finding frequent trajectories from the road network of a big city helps in summarizing the way the traffic behaves in the city. It can be very useful in city planning and traffic routing mechanisms, and may be used to suggest the best routes given the region, road, time of day, day of week, season, weather, and events etc. Other than the frequent patterns, even the events that are not so frequent, such as those observed when there is heavy snowfall, other extreme weather conditions, long traffic jams, accidents, etc. might actually follow a periodic occurrence, and hence might be useful to mine. This problem of mining the frequent patterns from road traffic data has been addressed in previous works using the context knowledge of the road network of the city. In this paper, we have developed a method to mine spatiotemporal periodic patterns in the traffic data and use these periodic behaviors to summarize the huge road network. The first step is to find periodic patterns from the speed data of individual road sensor stations, and use their periods to represent the station's periodic behavior using probability distribution matrices. Then, we use density-based clustering to cluster the sensors on the road network based on the similarities between their periodic behavior as well as their geographical distance, thus combining similar nodes to form a road network with larger but fewer nodes.
AB - The widespread use of road sensors has generated huge amount of traffic data, which can be mined and put to various different uses. Finding frequent trajectories from the road network of a big city helps in summarizing the way the traffic behaves in the city. It can be very useful in city planning and traffic routing mechanisms, and may be used to suggest the best routes given the region, road, time of day, day of week, season, weather, and events etc. Other than the frequent patterns, even the events that are not so frequent, such as those observed when there is heavy snowfall, other extreme weather conditions, long traffic jams, accidents, etc. might actually follow a periodic occurrence, and hence might be useful to mine. This problem of mining the frequent patterns from road traffic data has been addressed in previous works using the context knowledge of the road network of the city. In this paper, we have developed a method to mine spatiotemporal periodic patterns in the traffic data and use these periodic behaviors to summarize the huge road network. The first step is to find periodic patterns from the speed data of individual road sensor stations, and use their periods to represent the station's periodic behavior using probability distribution matrices. Then, we use density-based clustering to cluster the sensors on the road network based on the similarities between their periodic behavior as well as their geographical distance, thus combining similar nodes to form a road network with larger but fewer nodes.
KW - KL-divergence
KW - density-based clustering
KW - periodic patterns
KW - probability distribution matrices
KW - road network
KW - spatiotemporal data
KW - traffic data
UR - http://www.scopus.com/inward/record.url?scp=84884143854&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84884143854&partnerID=8YFLogxK
U2 - 10.1145/2505821.2505837
DO - 10.1145/2505821.2505837
M3 - Conference contribution
AN - SCOPUS:84884143854
SN - 9781450323314
T3 - Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining
BT - 2nd International Workshop on Urban Computing, UrbComp 2013 - Held in Conjunction with KDD 2013
T2 - 2nd ACM SIGKDD International Workshop on Urban Computing, UrbComp 2013 - Held in Conjunction with the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2013
Y2 - 11 August 2013 through 11 August 2013
ER -