Discovering patterns in spatial data using evolutionary programming

The problem of unsupervised learning of patterns in spatial data is addressed using evolutionary programming. Hyperbox clusters of two-dimensional data are evolved in light of a minimum description length (MDL) principle. A series of experiments of increasing complexity indicates that evolutionary programming can be useful in generating clusters for data when no a priori classification information is available. In some cases, the best evolved clusters have lower MDL scores than the clusters that would correspond to the true probability distribution of the data. Brief consideration is given to using other functions for clustering, such as hyperellipses.