MOGA-based fuzzy data mining with taxonomy

Transactions in real-world applications usually consist of quantitative values. Some fuzzy data mining approaches have thus been proposed for deriving linguistic rules from such transactions. Since membership functions may have a critical influence on the final mining results, several genetic-fuzzy mining approaches have been proposed for mining appropriate membership functions and fuzzy association rules at the same time. Most of them, however, focus on a single level and consider only one objective function. This paper proposes a multi-objective multi-level genetic-fuzzy mining (MOMLGFM) algorithm for mining a set of non-dominated membership functions for mining multi-level fuzzy association rules. The algorithm first encodes the membership functions of each item class (category) into a chromosome according to the given taxonomy. Two objective functions are then considered. The first one is the knowledge amount mined out at different levels, and the second one is the suitability of membership functions. The fitness value of each individual is then evaluated using these two objective functions. After the evolutionary process terminates, various sets of membership functions can be used for deriving multi-level fuzzy association rules according to decision-makers. Experimental results on the simulated and real datasets show the effectiveness of the proposed algorithm.