基于相邻关系的地理标识语言空间线对象离群检测算法

Algorithms for detecting outlier spatial lines based on adjacent relationship for geography mark-up language data

提出了两种基于相邻关系的地理标识语言空间线对象离群检测算法:DOL-AR1和DOL-AR2,定义了基于相邻关系的空间线对象之间的相异度,DOL-AR1将基于相邻关系的相异度作为空间线对象之间的距离度量准则,利用Density-based Spatial Clustering of Applications with Noise算法检测出离群的空间线对象.算法DOL-AR2以基于相邻关系的相异度为准则对空间线对象进行聚类,根据每个簇的离群因子,检测该簇是否离群.实验结果表明,算法DOL-AR1和算法DOL-AR2都能有效地检测出离群的线对象,本文对提出的两种离群检测算法的性能进行了比较,发现算法DOL-AR2的效率要高于算法DOL-AR1的效率.

Outlier detection is an important problem for data mining.Outlier detection is intended to discover unexpected,interesting and useful patterns of further analysis.Spatial outlier detection is aimed at detection of spatial objects different from other spatial objects in their spatial attributes or topological relationships.Now,only point objects without line or polygon objects are considered in the existing spatial outlier detection algorithms in which different degrees on topological relationships are not included.Algorithms DOL-AR1 and DOL-AR2 are presented here for detecting outlier lines based on adjacent relationship for GML data.Adjacent relations between each spatial line and other spatial objects are computed firstly.The difference degree between different lines that have the same type as the previous line on adjacent relationship is defined.In algorithm DOL-AR1,the difference degree is used as the standard of the distance between different lines.Density-based spatial clustering of applications with noise is used in algorithm DOL-AR1 in order to detect outlier spatial lines.In algorithm DOL-AR2,firstly,cluster spatial lines by the difference degree on adjacent relationship,then define the outlier factor of each cluster,the outlier factor of the cluster determines whether the cluster is 'outlier' or not.The experimental results show that algorithms DOL-AR1 and DOL-AR2 both can detect outlier lines based on adjacent relationship accurately and effectively.However,algorithm DOL-AR2 can run more effectively.