In this chapter, we are continuing our exploration of more complex clustering algorithms that can be employed in non-convex tasks (that is, where, for example, K-means fails to obtain both cohesion and separation. A classical example is represented by interlaced geometries). We are also going to show how to apply a density-based algorithm to a complex dataset and how to properly select hyperparameters and evaluate performances according to the desired result. In this way, a data scientist can be ready to face different kinds of problems, excluding the less valuable solutions and focusing only on the most promising ones.
In particular, we are going to discuss the following topics:
- Spectral clustering
- Mean shift
- Density-based Spatial Clustering of Applications with Noise (DBSCAN)
- Additional evaluation metrics: Calinski-Harabasz index and cluster instability ...
