• The clustering results can be radically different when the scaling parameters
of the algorithms are slightly modified or there is some noise perturbation
among clusters. We call such a susceptibility the sensitivity to parameter
tuning and noise.
• Most of these methods tend to assign medium similarity between the boundary
instances among clusters with different densities. Therefore they fail to
quantify local density well, which may result in poor manifold reconstruction
and undesirable clustering results.
• Most of the existing density-aware algorithms are only applicable on the Euclidean
space. Therefore their capabilities are significantly constrained in
handling today’s various types of data, such as social networks and text
datasets.