Finding Pareto-front Membership Functions in Fuzzy Data Mining

Abstract

Transactions with quantitative values are commonly seen in real-world applications. Fuzzy mining algorithms have thus been developed recently to induce linguistic knowledge from quantitative databases. In fuzzy data mining, the membership functions have a critical influence on the final mining results. How to effectively decide the membership functions in fuzzy data mining thus becomes very important. In the past, we proposed a fuzzy mining approach based on the Multi-Objective Genetic Algorithm (MOGA) to find the Pareto front of the desired membership functions. In this paper, we adopt a more sophisticated multi-objective approach, the SPEA2, to find the appropriate sets of membership functions for fuzzy data mining. Two objective functions are used to find the Pareto front. The first one is the suitability of membership functions and the second one is the total number of large 1-itemsets derived. Experimental comparisons of the proposed and the previous approaches are also made to show the effectiveness of the proposed approach in finding the Pareto-front membership functions.

A two-stage multi-objective genetic-fuzzy mining algorithm

“…Many MOEA-based GFM approaches have also been proposed with diverse objective functions [2,6,8,16]. Kaya proposed a multi-objective genetic algorithm based approach for discovering these optimized rules with three objective functions that are strongness (or support), interestingness (or confidence) and comprehensibility (or average number of fuzzy sets in the rule) [16].…”

“…Then, Kaya et al proposed another MOGA-based automated clustering approach for deciding the number of fuzzy sets and the fuzzy association rules with the two objective functions that are the number of large itemsets and the gain of time [2]. In [6,8], they also proposed MOGA-based approach for mining membership functions and fuzzy association rules with the given taxonomy and two objective functions that are the suitability of membership functions and the number of large itemsets. In the following, the descriptions of them are stated:…”

“…Many MOGA-based GFM approaches have also been proposed with diverse objective functions, e.g., number of large itemsets, number of rules, suitability of membership functions [2,6,8,16]. Although the derived non-dominated solutions could provide different chooses to users, they still meet the dilemma of how to take appropriate solution among them.…”

Rare-PEARs: A new multi objective evolutionary algorithm to mine rare and non-redundant quantitative association rules

“…Attributes of association rules can be placed in two 55 domains: the discrete and the continuous domain [4]. 56 Furthermore, association rules in the unsupervised domain can 57 be classified into two groups: categorical association rules versus 58 quantitative association rules [10,51] and frequent association 59 rules versus infrequent/rare rules [50]. There is also another type 60 of association rules, called class association rules.…”

“…We have developed our approach with three 132 motivations: 133 Ability of handling datasets with quantitative values. Previous 134 researches often work on binary or discretized values 135 [5,13,[58][59][60]. 136 Produce initial population by the aid of evolutionary random 137 permutation for all of sub-problems.…”