Signatures of past events in the Milky Way can be found in the form of structures in the position, velocity, and chemical abundance space. To study this, large amounts of multi-dimensional data already exists and numerous upcoming missions will create even larger and complex datasets. To efficiently and accurately analyze such datasets we develop a group finding algorithm which can work in a space of arbitrary number and type of dimensions. We develop a novel scheme which uses the idea of Shannon entropy to calculate a locally adaptive distance metric for each data point so as to extract maximum information from the data. We first apply this to the 2MASS data set (data with 2 angular positions and radial distance) and successfully identify some of the known structures in the form of tidal streams and dwarf galaxies and also predict some new structures. In order to understand the accretion history of the Milky Way we compare these results with synthetic surveys created out of simulated stellar halos (constructed within the LCDM cosmology). With an eye on future upcoming missions like GAIA and WFMOS; which will also have additional information in the form of proper motion, radial velocity and chemical abundances, we create similar synthetic surveys from our simulated stellar halos and analyze them. We show how adding the velocity and abundance information greatly improves the identification of structures. We also study the effect of observational errors on the identification of structures.