英语论文网

墨尔本网络作业本科3000字The traditional clustering algorithm

The traditional clustering algorithm based on grid defects are mainly embodied in:
1) in the grid partition in general use, not even the flexibility. For some, need to use small grid can be divided to describe it, if the problem space are classified and the size of the grid, then the uniform mesh grid cell number will be very much, sometimes even more than the number of data points, it will greatly increase the calculation time, If you are using large grid, although can increase speed, but clustering small ideal. Actually the inner regions of strong correlation with large size, can the grid, and the border region of small, in order to get accurate correlation with the description of small grid to divide. This method not only conforms to the characteristics of the data distribution, also reduced the number of grid.
2) not found any shape, because each class of data distribution is different, if judged according to the threshold, will cause missing parts or border area. Actually, belong to the same kind area is connected with certain similarity, and therefore can according to the similarity search through all of the unit with similar unit, and then to the adjacent unit for expansion, and find the class of arbitrary shape.
3) from the dataset can also find the Angle, global and local outlier
 
传统的基于网格的聚类算法的缺陷，主要体现在：1）在网格划分中普遍使用，甚至不灵活。对一些人来说，需要使用小网格可分为描述它，如果问题空间划分和网格的大小，然后均匀网格网格单元数量会很多，有时甚至超过了数据点的数量，这将大大增加了计算时间，如果你是用大网格，虽然可以提高速度，但集群的小理想。实际上，区域内强相关性大的尺寸，能网格，和边境地区小，以获得准确的相关描述小网格划分。这种方法不仅符合特征的数据分布，也减少了一些网格。2）没有发现任何形状，因为每一类数据的分布是不同的，如果根据判断阈值，将导致丢失部分或边境地区。实际上，属于同一类区域连接有一定的相似性，因此可以根据相似性搜索通过所有单元具有类似的单位，然后到邻近的单位进行扩建，并找到一类任意形状。3）从数据也可以找到角度，全局和局部异常