Dynamic local search for clustering with unknown number of clusters

Dynamic clustering problems can be solved by finding several clustering solutions with different number of clusters, and by choosing the one that minimizes a given evaluation function. This kind of brute force approach is general, but not very efficient. We propose a new dynamic local search that solves the number and location of the clusters jointly. The algorithm uses a set of basic operations, such as cluster addition, removal and swapping. The clustering is found by the combination of a trial-and-error approach of local search, and the local optimization capability of the generalized Lloyd algorithm. The algorithm finds the results 30 times faster than the brute force approach.