An inductive fuzzy class IFC regarding a predition target y is a fuzzy set of individuals i from a universe of discourse U defined by the fuzzy restriction i is likely a member of y.

IFC(y) := { i element of U | i is likely a member of y }

There are several ways to infer a membership function to IFC(y). In the mentioned publications, it is proposed to calcluate normalizations of likelihood comparisons.

The membership functions (MF) induced accordingly, can be used for data selection, visualization, and prediction in analytics. Correlations of MF with binary target indicators rank attributes by their relevance. 2D graphs of induced MF's provide an intuitive view into an attribute's target association. And finally, an inductive fuzzification of attributes using the induced MF's can improve predicitve performance of existing prediction algorithms.

IFC-Filter for Weka:

A Weka-implementation of the IFC-NLR machine learning algorithm can be downloaded here.

Publications:

• Kaufmann M., Graf C.: Inductive Fuzzy Classification for Analytic CRM. In: Meier A. & Donzé L. (Eds.), Handbook of Research in Fuzzy Methods for Marketing and CRM (2012) IGI Global Link
• Graf, C. (2010). Erweiterung des Data-Mining-Softwarepakets WEKA um induktive

  unscharfe Klassifikation (Master's Thesis). Department of Informatics, University of

  Fribourg, Switzerland.
• Kaufmann M.: An Inductive Approach to Fuzzy Marketing Analytics. The 13th IASTED International Conference on Artificial Intelligence and Soft Computing, September 2009, Palma de Mallorca, Spain, pp. 107 - 112
• Kaufmann M., Meier A.: An Inductive Fuzzy Classification Approach applied to Individual Marketing. The 28th North American Fuzzy Information Processing Society Annual Conference, June 2009, Cincinnati Ohio, USA 
  IEEE Link
• Kaufmann, M., Meier, A.: An Inductive Fuzzy Classification Approach applied to Individual Marketing, Internal Working Paper, University of Fribourg, 2008.
  Download Paper