An important problem in the area of data mining is the development of effective measures interestingness for ranking discovered knowledge. The five principles that we will refer to as the the Measure of Interestingness Principles provide a foundation for an intuitive understanding of the term "interestingess" when used within this context.

We examine functions f of m variable, f(n1,...,nn) where f denotes a general mesaure of diversity, m and each ni ( ni assumed to be non-zero), are as defined in the previous section, and f(n1,...,nn is a vector corresponding to the values in the derived Count attribute (or numeric measure attribute) for some arbitrary summary whose values are arranged in descending order such that n1 >= ... >= nm (except for discussions regarding ILorenz, which requires that the values be arranged in ascending order). Since the principles presented here are for ranking the interestingess of summaries generated from a single dataset, we assume that N is fixed.

P1 - Minimum Value Principle

Principle 1

P2 - Maximum Value Principle

principle 2

P3 - Skewness Principle

 principle 3

P4 - Permutation Invariance Principle

principle 4

P5 - Transfer Principle

principle 5

Measures Satisfying the 5 Principles

The following table identifies the measures that satisy the principles of interestingness. In the table the P1 to P5 columns describe the principles, and a measure that satisfies a principle in denoted by the bullet symbol

satisfy measure