An elliptical-shaped density-based classification algorithm for detection of entangled clusters

We present a density-based clustering method producing a covering of the dataset by ellipsoidal structures in order to detect possibly entangled clusters. We first introduce an unconstrained version of the algorithm which does not require any assumption on the number of clusters. Then a constrained version using a priori knowledge to improve the bare clustering is discussed. We evaluate the performance of our algorithm and several other well-known clustering methods using existing cluster validity techniques on randomly-generated bi-dimensional gaussian mixtures. Our simulation results show that both versions of our algorithm compare well with the reference algorithms according to the used metrics, foreseeing future improvements of our method.