DBSC: A dependency-based subspace clustering algorithm for high dimensional numerical datasets

We present a novel algorithm called DBSC, which finds subspace clusters in numerical datasets based on the concept of "dependency". This algorithm uses a depth-first search strategy to find out the maximal subspaces: a new dimension is added to current k-subspace and its validity as a (k+1)-subspace is evaluated. The clusters within those maximal subspaces are mined in a similar fashion as maximal subspace mining does. With the experiments on synthetic and real datasets, our algorithm is shown to be both effective and efficient for high dimensional datasets.