A Framework for Designing a Fuzzy Rule-Based Classifier
This paper is concerned with a general framework for designing a fuzzy rule-based classifier. Structure and parameters of the classifier are evolved through a two-stage genetic search. The classifier structure is constrained by a tree created using the ev
- PDF / 623,373 Bytes
- 12 Pages / 430 x 660 pts Page_size
- 101 Downloads / 237 Views
Department Electrical & Control Instrumentation, Kaunas University of Technology, Studentu 50, LT-51368, Kaunas, Lithuania 2 Intelligent Systems Laboratory, Halmstad University, Box 823, S-30118 Halmstad, Sweden [email protected], [email protected], {adas.gelzinis,marija.bacauskiene}@ktu.lt
Abstract. This paper is concerned with a general framework for designing a fuzzy rule-based classifier. Structure and parameters of the classifier are evolved through a two-stage genetic search. The classifier structure is constrained by a tree created using the evolving SOM tree algorithm. Salient input variables are specific for each fuzzy rule and are found during the genetic search process. It is shown through computer simulations of four real world problems that a large number of rules and input variables can be eliminated from the model without deteriorating the classification accuracy. Keywords: Classifier, Fuzzy rule, Genetic algorithm, Knowledge extraction, Variable selection.
1
Introduction
Neural networks and support vector machines are probably the most popular data classification techniques. However, classifiers based on these techniques are not transparent enough and are often considered as “black boxes”. The transparency is very important in some application areas, such as medical decision support or quality control. By contrast, fuzzy rule-based systems and fuzzy decision trees are known for their transparency and ability of accounting for uncertainty. ANFIS [1], fuzzy ARTMAP [2] are examples of the most prominent fuzzy logic-based systems. It is well known that designing of fuzzy rule-based systems in high dimensional spaces is rather problematic. However, there are many problems characterized by a small or moderate number of variables. Moreover, quite often high dimensional data vary in a much lower number of dimensions if
We acknowledge the support from the agency for international science and technology development programmes in Lithuania (COST IC0602). Corresponding author.
F. Rossi and A. Tsoukis (Eds.): ADT 2009, LNAI 5783, pp. 434–445, 2009. c Springer-Verlag Berlin Heidelberg 2009
A Framework for Designing a Fuzzy Rule-Based Classifier
435
compared to the dimensionality of an input space. System structure identification and parameter optimization are two main issues to consider when designing a fuzzy rule-based system [1]. Fuzzy partitioning, variable selection, and fuzzy reasoning are the tasks to be solved for identifying the system structure. Various approaches have been used for dealing with the two main fuzzy rule-based system design issues. The initial system structure, often termed as fuzzy partitioning, is usually identified through K-Means [3], Fuzzy C-Means [4], Learning Vector Quantization (LVQ) [5] or SOM-based clustering [6,7] as well as incremental clustering [8,9] or by constructing a decision tree [10,11]. Variable selection based on: the output sensitivity to the input change [12,13], the output sensitivity combined with the correlation between variables [6], Fisher’s
Data Loading...
