Implementation of the Model
Clustering is a type of unsupervised learning technique in which there are no explicit labels. Clustering is used to discover groups of data points in a dataset. A group or cluster is made up of members that are similar to each other but are collectively different from other clusters. A good clustering algorithm must have the ability to discover some or all hidden clusters in a dataset, should exhibit in cluster similarity but different clusters should be dissimilar or far from each other. The clustering algorithm should also be scalable to larger datasets and should be able to handle noisy data points and outliers.
There are two main categories of clustering algorithms - hierarchical clustering algorithms and partitive clustering algorithms. In hierarchical clustering, there is a clear relationship between discovered clusters. This can take the form of a hierarchy or order. Whereas in partitive algorithms, the relationship between clusters is not clear and it is sometimes referred to as having a flat structure. Clustering algorithms can also be seen from another perspective. Algorithms that allow a data point to belong to more than one cluster are known as soft clustering algorithms. In such a process, probabilities are assigned to each data point to indicate how likely it belongs to any particular cluster. Hard clustering algorithms on the other hand, require that data points only belong to exactly one cluster.
