The use of information and information gain in the analysis of attribute dependencies

This paper demonstrates the possible conclusions which can be drawn from an analysis of entropy and information. Because of its universality, entropy can be widely used in different subjects, especially in biomedicine. Based on simulated data the similarities and differences between the grouping of attributes and testing of their independencies are shown. It follows that a complete exploration of data sets requires both of these elements. A new concept introduced in this paper is that of normed information gain, allowing the use of any logarithm in the definition of entropy.