Clustering algorithms are available in the following categories:
One-level approach
A hierarchical method creates a hierarchical decomposition of a given set of data objects. The hierarchical method can be divided into condensation and splitting methods according to the formation of hierarchical decomposition.
Cohesion method: bottom-up. Start by forming a separate group for each object, and then merge the similar groups into the hierarchy until all the combinations are merged into one or a termination condition is met.
Splitting method: Top-down. Begins to place all objects in a cluster, each iteration, and the clusters split into smaller clusters until each object is in a cluster or satisfies a certain termination condition.
Two partitioning methods
Given n objects or data tuples of the database, the partitioning method constructs the K-partition of the data, each divided into a cluster, K N. given the divided array K to construct, the dividing party creates an initial partition. Then, iterative relocation techniques are used to try to improve the partitioning by moving objects between groups.
The typical partitioning method is: K-means and K Center point
2.1 K-means
Input
- K, number of clusters
- D, a DataSet containing N objects
Output
- A collection of k clusters
Process
- Arbitrary selection of K objects from D as the initial cluster center
- Repeat
Assigns each object to the most similar cluster based on the mean of the objects in the cluster;
Update cluster mean
Until clusters no longer change
K-means of clustering algorithm, K center point, hierarchical methods