Mining Fuzzy Association Rules: A General Model Based on Genetic Network Programming and its Applications

Abstract

The initiative of combining association rule mining with fuzzy set theory has been applied frequently in recent years [1][2][3][4][5]. The original idea comes from dealing with quantitative attributes in a database, where discretization of the quantitative values into intervals would lead to under or overestimation of the values that are near the borders. This is called the sharp boundary problem. Fuzzy sets can help us to overcome this problem by allowing different degrees of the membership, not only 1 and 0 treated by traditional methods. Attribute values can thereby be the members of more than one set and therefore give a more realistic view on such data. On the other hand, fuzzy set theory has been shown to be a very useful tool in association rule mining, because the mined rules can be expressed in linguistic terms, which are more natural and understandable for human beings. The linguistic representation is mainly useful when those discovered rules are presented to human experts for study. In this paper, a novel association rule mining approach that integrates the evolutionary optimization technique 'genetic network programming (GNP)' and fuzzy set theory has been proposed for mining interesting fuzzy rules from given quantitative data. The performance of our algorithm has been compared with other relevant algorithms and the experimental results show the advantages and effectiveness of the proposed model. 

Analysis of Various Interestingness Measures in Class Association Rule Mining

“…It should be noticed that the best coverage threshold (α) should be studied and the appropriate number of rules should be selected to achieve the best classification performance through the evaluation by the appropriate measure. In our simulations, we used an evolutionary algorithm based CARM [27], [28], [30] to study the effectiveness of interestingness measures, where multiple rules are used to predict the class labels. However, actually the framework of the proposed approach can be easily applied and testified in the other CARM methods.…”

“…The used method in this paper is called Genetic Network Programming (GNP) based CARM, where an evolutionary algorithm GNP is used to extract multiple CARs and an average matching degree mechanism is used to build the classification model. The effectiveness of this method has been proven in some previous research [27], [28]. Later, we will briefly introduce this method.…”

“…Note that the proposed approach mainly focuses on studying the selection of interestingness measures, therefore, any CARM method can be used, which will not be contrary to the aim of the topic. In this paper, we adopt a multi-rule based CARM method to study the topic, since the previous research has proved that multi-rule based algorithms can generally have better performance than single-rule based ones [9], [27], [28]. The used method in this paper is called Genetic Network Programming (GNP) based CARM, where an evolutionary algorithm GNP is used to extract multiple CARs and an average matching degree mechanism is used to build the classification model.…”

“…Since the rule extraction is not the key point of this topic, and actually any method can be used in the rule extraction step, which will not be contrary to the selection of measures, the details of GNP-EDA based CARM will not be shown in this paper. For GNP based CARM, the readers can refer to [10], [27], [28], and for GNP-EDA based CARM, the readers can refer to [29], [30]. In the following part, we just briefly describe the advantages of GNP based CARM and GNP-EDA based CARM.…”

“…The average matching degree mechanism has been applied to various problems, and its effectiveness has been proven by comparing it with several classical methods in the benchmark datasets, such as C4.5, CBA and CMAR [27], [28]. 5.…”