No abstract

In this paper, we propose a two-stage multi-objective fuzzy mining algorithm for dealing with linguistic knowledge discovery. In the first stage, the multi-objective genetic algorithm is used to derive a set of non-dominated membership functions (Pareto solutions) with two objective functions. In the second stage, the clustering technique is utilized to find representative solutions from the Pareto solutions. The representative solutions could be employed to mine fuzzy association rules according to the favorites of decision makers. Experiments on a simulation dataset are made and the results show the effectiveness of the proposed algorithm.

Mining association rules is an important technique in data analysis. Many approaches for rule analysis have been designed to address different problems. Among them, some works developed from multiobjective genetic algorithms derive a set of Pareto solutions, each of which contains a set of membership functions for fuzzy data mining from quantitative transactions with taxonomy. However, because more than one solution exists in a Pareto set, finding a method to determine the appropriate membership functions and combine them with useful knowledge for mining actionable patterns (such as fuzzy generalized association rules and fuzzy utility itemsets) is a useful research problem. Hence, this paper presents a post-analysisbased genetic-fuzzy mining (PA-GFM) framework for mining actionable patterns that involves two phases: membership-function mining and actionable pattern mining. In the first phase, an existing approach is utilized to derive the Pareto solutions with objective functions. In the second phase, a clustering technique using clustering attributes selected by the users is employed to group the Pareto solutions. The representative solution from each group is then exploited to mine actionable patterns based on the users' requirements. Experiments were conducted on both a simulated dataset and a real one to investigate the performance of the PA-GFM framework. INDEX TERMS Clustering algorithms, domain-driven data mining, fuzzy generalized association rules, fuzzy utility itemsets, multiobjective genetic algorithms.