Many methods have been proposed to mine fuzzy association rules from databases with crisp values in order to help decision-makers make good decisions and tackle new types of problems. However, most real-world problems present a certain degree of imprecision. Various studies have been proposed to mine fuzzy association rules from imprecise data but they assume that the membership functions are known in advance and it is not an easy task to know a priori the most appropriate fuzzy sets to cover the domains of the variables. In this paper, we propose FARLAT-LQD, a new fuzzy data-mining algorithm to obtain both suitable membership functions and useful fuzzy association rules from databases with a wide range of types of uncertain data. To accomplish this, first we perform a genetic learning of the membership functions based on the 3-tuples linguistic representation model to reduce the search space and to learn the most adequate context for each fuzzy partition, maximizing the fuzzy supports and the interpretability measure GM3M in order to preserve the semantic interpretability of the obtained membership functions. Moreover, we propose a new algorithm based on the Fuzzy Frequent Pattern-growth algorithm, called FFP-growth-LQD, to efficiently mine the fuzzy association rules from inaccurate data considering the learned membership functions in the genetic process. The results obtained over 3 databases of different sizes and kinds of imprecisions demonstrate the effectiveness of the proposed algorithm.
All Science Journal Classification (ASJC) codes
- Control and Systems Engineering
- Theoretical Computer Science
- Computer Science Applications
- Information Systems and Management
- Artificial Intelligence