A method and system are described for preparing a batch of data sets for a batch processing using a predetermined correlation encoding operation. The batch of data sets may be analyzed to determine a plurality of correlation parameters representing degrees of correlation between member data sets of said batch of data sets. The correlation parameters may be used to generate an optimized correlation dependency graph of the member sets. The optimized correlation dependency graph may be used for performing the predetermined correlation encoding operation. By optimizing the correlation dependencies between data sets for subsequent processing, e.g. data compression, of the data sets, the advantages of correlation encoding may be used, even for data sets which appear a priori to be unordered and only weakly correlated.