Clustering in machine learning is often essential. As the unstructured and unlabelled data is not always welcoming towards the deployment of automation tools. The clustering of data is done generally for unlabelled data. The process is concerned with the grouping of similar data. However, the word similarity is not very appropriate in this case, as based on other traits similar units can be placed in different groups. Clustering is an essential method when it comes to data analytics and the deployment of machine learning tools. In the case of commerce, bioinformatics, computer vision, image processing, and data comparison, clustering is one of the most used data analytics methods. Based on factors like origins and implications, data clustering can be performed following four major protocols.

- Connectivity based or hierarchical clustering
- Centroid dependent clustering
- Density clustering
- Distribution based clustering

These protocols are deployed based on the scenario at hand and the demands of an organization. This article will discuss these four major clustering techniques and explain their utility in the process.

**Hierarchical cluster analysis **

In this particular method of cluster analysis, a data set is considered as a family of all the clusters that are included in the data set. The output is a dendrogram, depicting the hierarchical relationship between different clusters.

To understand hierarchical clustering in detail, we must look at an example. Let there be 5 (A, B, C, D, E)different data clusters in a data set. To perform hierarchical clustering these clusters must have some kind of similarities so that they can be considered a family of clusters. Now, let A and E be related more closely than they are with the others and the same is true for B and C. Now these clusters can be merged into AE and BC clusters. And the isolated cluster D can then be taken up by any of these two new clusters in accordance with the degree of similarity. At the end of this loop, a giant cluster will be formed ( ABCDE) and if we try to depict the construction of the same in a dendrogram, it will be called hierarchical cluster analysis.

**Centroid dependent clustering**

The centroid in a cluster can occur naturally, or it can be a hypothetical point representing the mean value of all the other points in a data cluster. In order to arrange a set of units into a cluster, the centroid is taken randomly at first. Then the nearest units, in terms of one or more than one similarity are clustered in a data cluster. A data analyst in need of multiple clusters will take more than one cluster across the data set and cluster nearest units into multiple clusters. Among the clustering techniques, this one is perhaps the most deployed and utilized due to the versatile nature of the protocol. In a single set, multiple genres of clustering can be done thanks to the light of different genres of similarities between units. After a centroid-dependent clustering is done each unit in a cluster is expected to be similar to other members of the same cluster to some degree. And units from different clusters are expected to be different in the light of one or a few criteria. After the clustering is done, the imaginary or arbitrary centroid is validated or the original centroid is identified.

**Density-dependent clustering**

Density-dependent clustering is most commonly seen in the case of analyzing large-scale structure surveys. In this cluster analysis paradigm, a region across the data set is identified where the values seem to cluster naturally. And these densely occupied areas of value must be separated areas with little value to speak of. This differential distribution is exploited in density cluster analysis. instead of one, multiple points are assumed across a data set, and values nearest to them are clustered across a region of relatively more units. In the case of data sets ubiquitously occupied with values, it is hard to employ density dependent clustering. And in those cases centroid and distribution clustering techniques are more preferred.

**Distribution dependent clustering**

Distribution-dependent clustering involves the deployment of distribution models like the gaussian model and the normal distribution model. By analysis of this distribution, the authenticity of a naturally obtained unstructured data set can be evaluated. The distribution models are also among one the most used clustering techniques and are being used in a plethora of scenarios. But distribution based analysis suffers due to the chances of overlapping features and values. Thus is not applicable ubiquitously.