Mining Fuzzy Moving Object Clusters
-
Upload
nhathai-phan -
Category
Data & Analytics
-
view
139 -
download
1
Transcript of Mining Fuzzy Moving Object Clusters
1
Mining Fuzzy Moving Object Clusters
Phan Nhat Hai, Dino Ienco
Pascal Poncelet, Maguelonne Teissiere
The 8th International Conference on Advanced Data Mining and Applications (ADMA 2012)
2
Outline
Background and Motivation
Fuzzy-Closed Swarm Definition
f-CS Miner Algorithm
Experimental Results
3
Trajectory Data
Spatio-Temporal Data. Represented as a set of points, located in space and time.
T=(x1,y1, t1), …, (xn, yn, tn) position in space at time ti was (xi, yi).
4
Mining Movement Patterns
Clustering:Group together similar trajectories.
Consecutive timestamps:Flock [GIS 2006], moving cluster [TKDE 2007], convoy [PVLDB 2008, SSDBM 2010], k-Star [SDM 2009]
Non-consecutive timestamps:Closed swarm [PVLDB 2010]
rGpattern [GIS 2012]
5
Motivation
Completely relaxing consecutive time constraint:
Generate a large number of patterns
Many of them are extraneous
6
Motivation
Propose:Fuzzy time gap
Fuzzy time gap participation index
Questions:Which is relevant time gap?
When the pattern extension should be stopped?
Pattern:o1 and o2 are moving together from {A>B>C} with 60% weak, 20% medium and 20% strong time gaps
7
Outline
Background and Motivation
Fuzzy-Closed Swarm Definition
f-CS Miner Algorithm
Experimental Results
12
Outline
Background and Motivation
f-Closed Swarm Definition
f-CS Miner Algorithm
Experimental Results
13
Straight Forward Approach
Extract all closed swarmsObjectGrowth [PVLDB 2010]
Post-processing to obtain f-closed swarms
Expensive task:Search space: O(2|ODB| x 2|TDB|)
New data is always availableExecute again and again the algorithms on the whole database, i.e. including existing data and new data
17
Outline
Background and Motivation
f-Closed Swarm Definition
f-CS Miner Algorithm
Experimental Results
18
Datasets
Datasets: Swainsoni dataset
43 objects, 4225 different timestamps, gathered from July 1995 to June 1998.
Buffalo dataset165 buffalos, gathered from year 2000 to year 2006.
Synthetic data: 500 objects - 10,000 timestamps.
Hardest condition
LCM algorithm [ICDM FIMI 2004]
22
Conclusion and Future Directions
Propose concept of f-Closed swarm
Propose f-CS Miner
Demonstration system
Extension directions:Mining top-K representative f-closed swarms
Apply on other patterns, i.e. rGpattern [GIS 2012]
23
THANK YOU FOR YOUR ATTENTION Question and Answer
[email protected]@[email protected]@teledetection.fr
The 8th International Conference on Advanced Data Mining and Applications (ADMA 2012)
Demo website: http://www.lirmm.fr/~phan/fcsminer.jsp