Genetic algorithm for data clustering based on SV-criterion

We consider the problem of data clustering in complex conditions: distributions of features within clusters differ significantly; the available data sample contains invalid examples; feature set may be incomplete for an unambiguous separation of clusters. We advance the previously proposed SV-criterion and offer its generalization for correlated features. A genetic algorithm using the proposed criterion is developed. The algorithm is tested on simulated data. Research has shown the possibility of determining the true number of clusters in the data set.