A modified electromagnetic-like mechanism for rough set attribute reduction

چکیده مقاله

This paper presents a modified electromagnetic-like mechanism (MEM) designed specifically for rough set attribute reduction, which is a critical preprocessing step in data mining and classification tasks. Attribute reduction aims to remove redundant and irrelevant attributes while preserving the essential information of the original dataset. The authors adapt a continuous optimization algorithm, the electromagnetic-like mechanism (EM), to solve this discrete optimization problem by introducing a novel discretization function that maps continuous values to binary representations of attribute subsets. This marks the first application of EM to rough set-based attribute reduction.

The proposed MEM incorporates several key enhancements over the standard EM algorithm. It embeds two memory mechanisms to retain information from previous iterations, improving the algorithm's ability to escape local optima and refine the search process. Additionally, a tailored fitness function is developed based on rough set dependency degree to evaluate the quality of attribute subsets. The algorithm operates through an attraction-repulsion framework, where solutions with better fitness attract others, while poorer solutions repel, guiding the population toward minimal reducts.

Experimental validation is conducted using several well-known UCI datasets, comparing MEM against standard EM, genetic algorithm (GA), and particle swarm optimization (PSO). The results demonstrate that MEM consistently achieves competitive or superior reduction sizes across multiple datasets, often identifying minimal reducts with fewer iterations. Notably, MEM shows improved convergence speed and reduced sensitivity to parameter tuning compared to other meta-heuristic approaches.

The study concludes that MEM offers a balanced and effective method for attribute reduction, particularly in scenarios where processing time is a priority. The algorithm successfully combines the global search capabilities of EM with problem-specific adaptations for rough set theory. Future work may focus on further optimizing parameter settings using fuzzy rules or extending the approach to other combinatorial optimization problems in data mining.