CAJ | 학술논문

为了检测空间数据集中存在的离群区域, 提出一种基于裁边策略的空间离群区域检测算法。首先利用Delaunay三角网格确定空间邻接关系, 根据非空间属性描述邻接节点间的差异性; 然后反复裁去最大权边, 且并发地检测离群区域, 直到发现足够多离群点。实验结果表明, 该算法能有效检测离群区域, 并且准确给出局部离群性, 克服了普通算法中易受坏邻居干扰和区域缺乏原子性的局限。
위료검측공간수거집중존재적리군구역, 제출일충기우재변책략적공간리군구역검측산법。수선이용Delaunay삼각망격학정공간린접관계, 근거비공간속성묘술린접절점간적차이성; 연후반복재거최대권변, 차병발지검측리군구역, 직도발현족구다리군점。실험결과표명, 해산법능유효검측리군구역, 병차준학급출국부리군성, 극복료보통산법중역수배린거간우화구역결핍원자성적국한。
In order to detect the existed outlier regions in spatial data sets, this paper proposed a cutting-long-edges-based approach. It used Delaunay triangle mesh to construct adjacency relationships. Then it according to non-spatial attributes addressed the difference between nodes. Finaly it iteratively cut long edges, and concurrently detected anomalous regions, and it found till enough outlier. Experiments demonstrate that out approach can effectively discover anomalous regions and the abnormality measure can reflect real local difference. Especially, the approach overcomes the general drawbacks such as the disturbance of bad neighbors and the lack of atomicity of regions.