Abstract : Rough sets effectively handle discrete attributes for rule mining but struggle with continuous, fuzzy data. To address this, we propose particle swarm optimization (PSO)-based fuzzy c-means (FCM) and fuzzy rough set (FRS) approaches. PSO optimizes FCM cluster centers, improving fuzzification of continuous attributes over traditional methods. Then, instead of an equivalence relation in rough set, a fuzzy similarity relation is used and an inductive learning method based on FRS is introduced to generate rules. The method is evaluated against the fuzzy concept learning system (FCLS), Kohonen’s self-organizing map (SOM)-based FRS, and FCM-based FRS approaches using a case study on salary prediction. The results show superior performance of the proposed approaches: fitness values with 23%~33% improvement, number of rules with 56% reduction, and 100% coverage rate and accuracy. The framework’s efficiency in managing fuzzy, continuous data underscores its potential for real-world decision systems, offering a balanced trade-off between interpretability and precision.

Keyword : Data mining algorithms, fuzzy clustering, fuzzy rough set, particle swarm optimization, rule mining.