Mining spatial trajectories using non-parametric density functions
MLDM'11 Proceedings of the 7th international conference on Machine learning and data mining in pattern recognition
Hi-index | 0.00 |
Trajectory clustering is attractive for the task of class identification in spatial database. Existing trajectory clustering algorithm TRCLUS uses global parameters to discover common trajectories. However, it can not discover small and dense clusters and be sensitive to two input parameters. Based on the partition-and-group framework, we propose a simple but effective trajectory clustering algorithm based on symmetric neighborhood named BSNTC, which needs only one input parameter which eases the sensitivity of parameters in a certain extent. The proposed measure considers both neighbors and reverse neighbors of trajectories when estimating its density distribution. Also, we use an accumulator without calculating influence outlier of each trajectory to reduce the computation cost and corresponding storage cost. A comprehensive performance evaluation and analysis shows that our method is not only efficient in the computation but also effective in arbitrary shape and different densities trajectory databases.