In this chapter, we are going to introduce some fundamental clustering algorithms, discussing both their strengths and weaknesses. The field of unsupervised learning, as well as any other machine learning approach, must be always based on the concept of Occam's razor. Simplicity must always be preferred when performance meets the requirements. However, in this case, the ground truth can be unknown. When a clustering algorithm is adopted as an exploratory tool, we can only assume that the dataset represents a precise data generating process. If this assumption is correct, the best strategy is to determine the number of clusters to maximize the internal cohesion (denseness) and the external separation. This means that we expect to find blobs (or isles) whose samples share some common and partially unique features.
In particular, the algorithms we are going...
