Pattern Recognition, 4th Edition

Grid-based Subspace Clustering Algorithms (GBSCAs)

The main strategy adopted by these algorithms consists of the following steps:

(a)identify the subspaces of the feature space that are likely to contain clusters, (b)determine the clusters lying in each of these subspaces, and (c) obtain descriptions of the resulting clusters.

The algorithms of this family apply an l-dimensional grid on the feature space and identify the subspaces that are likely to contain clusters, based on the k-dimensional units (boxes) (k ≤ l) defined by the grid. However, the consideration of all possible subspaces becomes infeasible, especially when high-dimensional data sets are considered. To solve this problem, the algorithms establish certain criteria that are ...