Hierarchical clustering is used when a underlying application needs a hierarchy. Such kind of a cluster produces high quality clusters. However there are many problems and limitations with the clustering technique which are as follows :
Ability to handle different cluster sizes
Hierarchical clustering can face problems in handling the relative sizes of the clusters that are merged. There are two approaches : weighted, which treats all clusters equally and unweighted, which takes number of points in each cluster into account. Here the number of weighted and unweighted refers to the data points not the clusters. In other words treating clusters of unequal size equally gives different weights to the points in different clusters, while taking cluster size into account gives points in different clusters the same weight. This is one major limitation of the hierarchical clustering algorithm.
Merging Decisions are Final
Agglomerative hierarchial clustering algorithms tend to make good local decisions about combining two clusters since they can use information about the pairwise similarity of all points. However, once a decision is made to merge two clusters, it cannot be undone at a later time. This approach prevents a local optimization criterion from becoming a global optimization criterion.
Cost Factor
Hierarchial clustering algorithm is expensive in terms of its computational and storage requirements. The fact that all merges are final can also cause trouble for noisy, high dimensional data such as document data.
In turn these problems can be addressed to some degree by first partially clustering the data using another technique called K-means.
No comments:
Post a Comment