Hands-On Unsupervised Learning with Python

DBSCAN

DBSCAN is another clustering algorithm based on a density estimation of the dataset. However, contrary to mean shift, there is no direct reference to the data generating process. In this case, in fact, the process builds the relationships between samples with a bottom-up analysis, starting from the general assumption that X is made up of high-density regions (blobs) separated by low-density ones. Hence, DBSCAN not only requires the maximum separation constraint, but it enforces such a condition in order to determine the boundaries of the clusters. Moreover, this algorithm doesn't allow specifying the desired number of clusters, which is a consequence of the structure of X, but, analogously to mean shift, it's possible to control the ...