Research For Spatial-temporal Outlier Detection

Spatio-temporal data mining is meant to extract interesting spatio-temporalpatterns and characteristics, spatio-temporal and non-spatio-temporal generalrelationship, and other knowledge implied in the spatial-temporal database. Outlierdetection is used to find abnormal patterns in small proportion data. Spatio-temporaloutlier detection is one of important parts of spatio-temporal data mining, and is theextension of outlier detection on spatio-temporal data, which is used to findspatio-temporal objects or patterns whose non-spatio-temporal attributes deviate so馝馝馝馝趑死(?)□塨攭□ □(?)□伀(?)倀(?)愀□(?)(?)□塨攭□ □(?)傈(?) □伀(?)倀(?)愀heir spatio-temporal neighborhood. In recent years, gigantic spatio-temporal dataoccupying certain spatial location are continuously updating, such as remote sensingdata, there must be many exceptional data and patterns which behave some unusualproperties, spatial-temporal outlier detection has already been a significant subject inoutl Teheletertidne fiqhbblems analyzed about spatial-temporal outlier detection, whichare how to define spatio-temporal neighborhood and how to judge spatio-temporaldata's outlying degree. We made some research works on temporal outlier detectionand spatial outlier detection, which are extending point of spatio-temporal outlierdetection. On the temporal outlier detection, a new segment method of temporal datawas proposed, which divided temporal data by combining important point with theirchanging properties, and can be used for effective temporal outlier mining. On thespatial outlier detection, a new measurement of spatial outlying degree, spatial localoutlier coefficient (SLOC) was proposed, which combined neighborhood distancedenoting dependency between spatial objects with neighborhood distributingproperties. On the definition of spatio-temporal neighborhood, we summed up threespatial neighbor relations based on meet, connectivity and Voronoi, and two temporalneighbor relations based on fluctuation and cycle. Combining spatial neighborrelations with temporal neighbor relations, we gained formalized definition ofneighborhood of spatial-temporal object.On the basis of the above study, we converted our focuses on spatio-temporaloutlying pattern detection. In this paper, spatio-temporal patterns, which are temporalsegment of spatio-temporal data occupying same spatial location, were introduced asbasic elements of spatio-temporal detection. Based on the concept of neighborhood ofspatio-temporal object, we defined the neighborhood of spatio-temporal pattern. Fromthe spatial outlier detection algorithm(SLOC),we made temporal extension andpresent a robust algorithm of spatio-temporal outlying pattern detection based onspatial temporal local outlier coefficient (STLOC).The experimental results show thatour algorithm can effectively detect spatio-temporal patterns.