On data analysis and variable selection: the minimum entropy analysis

In this work, we present a minimum entropy analysis scheme for variable selection and preliminary data analysis. The variable selection can be achieved by the increasing preference of variables. We show such a preference to has a unqiue form, which is given by the entropy of models associated with variables. Evaluating the entropy provides a complete ranking scheme of variables. This scheme not only indicates preferred variables but also may reveal the system's nature and properties. We illustrate the proposed scheme to analyze a set of geological data for three carbonate rock units in Texas and Oklahoma, and compare to the discriminant function analysis. The result suggests this scheme to provide a quick and robust analysis, and the use in data analysis is promising.