Clustering

Clustering :

Unlike Supervised learning model, the Unsupervised model of machine learning has no predefined groups under which you can distribute your data. You can find these groupings through clustering.

As you can  see in the image below, the points are shown dots. These data points do not have labels based on which you can differentiate them. You don't know anything about this data. So, now the question is can you find out any structure in this data? This problem can be solved using clustering technique.

Clustering will divide this entire data set under different labels, with similar data points into one cluster as shown above. It is used as very powerful technique for exploratory descriptive analysis.

Here, the clustering technique has partitioned the entire data points into two clusters. The data points with in a cluster are similar to each other, but different from other clusters. For instance, you have the data set on symptoms of patients. Now you can find out the name of a particular disease based on these symptoms.

So, basically clustering partitions the data set with similarities into different groups which can act as a base for further analysis. Let's look at different ways to achieve these clusters.

Exclusive clustering:

In exclusive clustering, an item belongs to exclusively to one cluster, not several. In the image you see that data belonging to cluster 0 doesn't  belonging to cluster 1 or cluster 2. K- means clustering is a type of exclusive clustering.

Overlapping Clustering:

Here, an item can belong to multiple clusters with different degree of association among each cluster. Fuzzy c-means clustering is based on overlapping clustering.

Hierarchical clustering:

In hierarchical clustering, the clusters are not formed in a single step rather it fallows series of partitions to come up with final clusters. It looks like a tree as visible in the image.

While implementing any algorithm, computational speed and efficiency becomes a very important parameter for end results.