1

Clustering Methods

Student’s name

Institution affiliation

Professor’s name

Course name

Date

Clustering Methods

Centroid-based Clustering: K-means

K-means clustering is an unsupervised learning algorithm that requires data to be

categorized in a pre-defined number of clusters. The algorithm starts at a random point with

some centroids, and then all data points are assigned to the closest centroid based on measures

such as Euclidean distance. In later iterations, the centroid is recalculated as the mean of the

points assigned to that cluster, and points are reassigned based on this new centroid. This

proceeding will continue till the centroids stop being unstable or the maximum number of

iterations is over.

Connectivity-based Clustering: Hierarchical Clustering

Hierarchical clustering can be constructed by the divisive method (dividing clusters from

the top) or by the agglomerative method (merging clusters from the bottom). In the

agglomerative perspective, all data points begin as individual clusters that are merged pairwise

(such as single-linkage, complete-linkage, or average-linkage) as one moves up the hierarchical

structure based on the linkage criterion.

Density-based Clustering: DBSCAN

DBSCAN is designed to locate "clusters" crowded near a low-density region. It starts

with an arbitrary point, and if this point has a certain number of neighborhoods within a

particular radius, it will be a cluster (Bhattacharjee & Mitra, 2021). The neighboring points

within the radius should be investigated and included in the cluster because the clustering

process will continue until each density-connected cluster is determined.

3

References

Bhattacharjee, P., & Mitra, P. (2021). A survey of density-based clustering algorithms. Frontiers of Computer Science, 15, 1-27. https://link.springer.com/article/10.1007/s11704-019-9059-3