A GENERIC GENETIC PROGRAMMING(GGP) TO GENERATE AUTOMATICALLY DEFINED FUNCTIONS (ADF's) : application in datamining and information retrieval

Abstract

One of the main and fundamental tasks of data mining is the automatic induction of classification rules from a set of examples and observations . A variety of methods performing this task have been proposed and many comparative studies have been carried out in this field ; however the main common feature between these methods is that they are designed by humans, there have been a successful attempt to automatically design such methods using GGP. In this work we describe a different system that can evolve com plete java program codes representing rule induction algorithms using the grammatical evolution technique that governs a Backus Naur Form gram- mar definition mapping to a program, in this system we will use as inputs to the mapper along with the Backus Naur Form grammar, integer strings representing potential solutions resulting from the initialiser component and Weka building blocks to facilitate the induction process and shorten the induced programs . Rule induction algorithms evolved using this system can be of different complexities. Data in real world applications are in most cases linguistic information that is ambiguous and uncertain. Hence, such data should be handled by fuzzy set representation schemes to increase expressiveness and comprehensiveness. Moreover, mining these data requires ways to generate automatically useful information/knowledge through a set of fuzzy rules. In this work we propose an extension of this system is also proposed and tested. The new system is called FuAGGE which stands for Fuzzy Automatic Generator Genetic Expression. FuAGGE was conceived in a way that it can evolve rule based classifiers that generate fuzzy rule sets. FuAGGE uses a grammar based evolutionary technique. The grammar is expressed in the Backus Naur Form (BNF) and represents a fuzzy set covering method. The grammar is mapped into programs that are themselves implementations of fuzzy rule induction algorithms. FuAGGE has also been tested on a ix benchmark of well-known datasets, and experimental results prove the efficiency of the proposed method. It is shown through comparison that our method outperforms most recent and similar, manual techniques. The system is able to generate rule induction algorithms specialized to specific domains, for example medical or biological data. It produces efficient rule models and achieves more accurate classification.