Iterative evolutionary subspace clustering

We propose in this paper a subspace clustering in high dimensional datasets using an iterative evolutionary algorithm. The evolutionary algorithm offers an original alternative to solve the problem of selecting subspace to deal with complex data structures in different subspaces. These subspaces that shows specific data structures are chosen iteratively. This step allows us to cluster data and subspace simultaneously. The evolutionary approach evaluates a subset of dimensions with a measure to find the best data cluster and then repeat the process for several subsets selected iteratively by the evolutionary algorithm. Experiments and comparisons on data sets with a large number of measurements show the effectiveness of the proposed approach.