In this section, we will focus on the most fundamental issue of clustering algorithms, which is discovering the number of clusters in a dataset – there is no definitive answer. However, not all clustering algorithms require a predefined number of clusters. In hierarchical and DBSCAN clustering, there is no need to define the number of clusters, but in k-means, k-medoids, and spectral clustering, we need to define the number of clusters. Selecting the right value for the number of clusters is tricky, so let's look at a couple of the methods for determining the best number of clusters:
- The elbow method
- The silhouette method
Let's look at these methods in detail.
