Outlier detection in spatial data using the m-SNN algorithm
Outlier detection is an important topic in data analysis because of its applications to numerous domains. Its application to spatial data, and in particular spatial distribution in path distributions, has recently attracted much interest. This recent trend can be seen as a reflection of the massive...
| Main Author: | |
|---|---|
| Format: | Others |
| Published: |
DigitalCommons@Robert W. Woodruff Library, Atlanta University Center
2013
|
| Subjects: | |
| Online Access: | http://digitalcommons.auctr.edu/dissertations/1299 http://digitalcommons.auctr.edu/cgi/viewcontent.cgi?article=2633&context=dissertations |
| Summary: | Outlier detection is an important topic in data analysis because of its applications to numerous domains. Its application to spatial data, and in particular spatial distribution in path distributions, has recently attracted much interest. This recent trend can be seen as a reflection of the massive amounts of spatial data being gathered through mobile devices, sensors and social networks. In this thesis we propose a nearest neighbor distance based method the Modified-Shared Nearest Neighbor outlier detection (m-SNN) developed for outlier detection in spatial domains. We modify the SNN technique for use in outlier detection, and compare our approach with the widely used outlier detection technique, the LOF Algorithm and a base Gaussian approach. It is seen that the m-SNN compares well with the LOF in simple spatial data distributions and outperforms it in more complex distributions. Experimental results of using buoy data to track the path of a hurricane are also shown. |
|---|