Constructing Accurate Fuzzy Rule-Based Classification Systems Using Apriori Principles and Rule-Weighting
A fuzzy rule-based classification system (FRBCS) is one of the most popular approaches used in pattern classification problems. One advantage of a fuzzy rule-based system is its interpretability. However, we’re faced with some challenges when generating t
- PDF / 708,508 Bytes
- 10 Pages / 430 x 660 pts Page_size
- 62 Downloads / 187 Views
Abstract. A fuzzy rule-based classification system (FRBCS) is one of the most popular approaches used in pattern classification problems. One advantage of a fuzzy rule-based system is its interpretability. However, we're faced with some challenges when generating the rule-base. In high dimensional problems, we can not generate every possible rule with respect to all antecedent combinations. In this paper, by making the use of some data mining concepts, we propose a method for rule generation, which can result in a rule-base containing rules of different lengths. As the next phase, we use rule-weight as a simple mechanism to tune the classifier and propose a new method of ruleweight specification for this purpose. Through computer simulations on some data sets from UCI repository, we show that the proposed scheme achieves better prediction accuracy compared with other fuzzy and non-fuzzy rule-based classification systems proposed in the past. Keywords: Pattern classification, fuzzy systems, data mining, rule weighting.
1 Introduction Fuzzy rule-based systems have been widely used on control problems [1,2,3]. One key feature of fuzzy rule-based systems is their comprehensibility because each fuzzy rule is linguistically interpretable. Recently, fuzzy rule-based systems have been applied successfully on classification problems [4,5,6]. The interest in using fuzzy rule-based classification systems (FRBCS) arises from the fact that those systems consider both accuracy and comprehensibility of the classification result at the same time. In fact, error minimization and comprehensibility maximization are two conflicting objectives of these kinds of classification systems and the trade off between these two objectives has been discussed in some recent studies [7,8,9,10]. Basic idea for designing a FRBCS is to automatically generate fuzzy rules from numeric data (i.e., a number of pre-labeled training examples). Hence, rule-base construction for a classification problem always has been a challenging part of it. In this paper, a novel approach for generating a set of candidate rules of each class is presented using data mining principles in which the number of generated rules is H. Yin et al. (Eds.): IDEAL 2007, LNCS 4881, pp. 547–556, 2007. © Springer-Verlag Berlin Heidelberg 2007
548
S.M. Fakhrahmad, A. Zare, and M.Z. Jahromi
reduced dramatically. A compact rule-base is then constructed by selecting a specified number of candidate rules from each class (using a selection metric). In many studies, antecedent fuzzy sets were generated and tuned by numerical input data for rule-base construction to improve the classification accuracy of FRBCSs. As shown in [11,12], the modification of the membership functions of antecedent fuzzy sets can be replaced by rule weight specification to some extent. Since, the adjustment of membership functions may degrade the interpretability of a FRBCS. In this paper, a learning algorithm is proposed to adjust the weights of the rules (existing in the rule-base) by the training data. This met
Data Loading...
