基于粗粒度单元的离群点检测算法研究

对现有的基于单元的算法进行改进，利用KNN算法思想得到距离与比例参数的合理先验值，以加快离群点检测的收敛速度；同时通过扩大单元粒度，减少了单元区域查询次数与算法的空间复杂度，从而在整体上提高了离群点的检测效率。通过实验，验证了改进后算法的可行性，同时比较了其与原算法在不同参数下的性能优劣。

Researches on Improvement Cell-based Outlier Detection Algorithm

With the algorithm proposed, the original cell-based algorithm is improved by expanding the granularity of the cell. The process of outlier detection is optimized by giving a more reasonable initial parameter using KNN method. Meanwhile,the outliers are looked up in the expanded cell to reduce the amount of time and searching involved. The experimental result verified the feasibility and reliability of the improved algorithm. Meanwhile, the performance indexes of improved and original algorithm were compared through the experiment.