Evolving Fuzzy Classifiction Rules

Renu Bala, Saroj Ratnoo

Abstract


Evolutionary Algorithms (EAs), being a robust and adaptive search methods, have found extensive applications in various tasks of Knowledge Discovery. Several usages of Genetic Algorithms (GAs) for mining classification rules have shown potentially useful results. The disadvantage of the simple Rule Based Classification Systems is that they involve sharp cutoffs for distribution and hence are unable to deal with uncertainty and vagueness imperative to decision making situations. Fuzzy logic is a precise logic of imprecision and approximate reasoning. More specifically, fuzzy logic has capability to reason in an environment of imprecision, uncertainty and incompleteness of information. Therefore, Fuzzy Systems became very popular in the domain of control applications and expert systems. Regardless of the great success of fuzzy systems, currently, there has been an increasing interest to augment fuzzy systems with learning and adaptation capabilities. This necessitated the integration of EAs with Fuzzy Logic.  This paper presents an extensive review on evolving Fuzzy Classification Rules (FCRs) employing Genetic algorithms or Genetic Programming. Learning FCRs involves learning of Data Base (DB) that contains the definitions of linguistic terms, fuzzy membership functions or fuzzy partitions etc. and Rule Base (RB) which contains Fuzzy Classification Rules most often in the form of high level symbolic IF-Then rules consisting of antecedent and consequents with fuzzy constructs. As the problem of rule mining is multi-objective, the application of Multi Objective GAs (MOGAs) is gaining ground to deal with conflicting criteria like accuracy and comprehensibility of the discovered rule set. Hence the recent implementations of MOGA in the area of rule mining/fuzzy rule mining have also been discussed. In the end, some unresolved problems are taken up and there is an attempt to lay down some directions for future research.


Full Text:

PDF

Refbacks

  • There are currently no refbacks.


Subscribe to Print Journals

 IJAIKD is currently Indexed By   


 http://rgjournals.com/public/site/images/mittalberi/scholar_logo_lg_2011.gif  Journal Seek