| Recent proliferation of low-cost and lightweight GPS tracking devices led to a large increase in the amounts of collected mobility data. The rapidly emerging field of location-based services requires accurate and informative knowledge mining from these large quantities of data. One such mobility knowledge mining task is trajectory clustering, where one tries to find paths that have been travelled frequently. Most existing trajectory clustering techniques do not discuss cleaning the data before applying a clustering algorithm. Since "noisy" data can have a significant effect on the clustering process, preprocessing such trajectory data will likely improve trajectory clustering results. In this thesis, we present a trajectory data cleaning framework, which consists of four steps: Outlier Detection, Stop Detection, Interpolation and Map Matching. We evaluate our framework using popular clustering algorithms and distance functions, and show that our proposed preprocessing (cleaning) framework indeed does improve the quality of obtained clusters. |
