نوع مقاله : مقاله پژوهشی
عنوان مقاله English
نویسندگان English
In this study, we present a novel algorithm that enhances the noise resilience of Learning Classifier Systems through the integration of cellular automata dynamics. The proposed framework is assessed using both synthetic and real-world datasets with significant noise contamination, conditions that typically impair classification accuracy, slow evolutionary convergence, and reduce the interpretability of rule-based models in conventional Learning Classifier Systems architectures. In the proposed design, classifier rules are encoded as cells that perform localized interactions within a structured neighborhood defined by either the Von Neumann or Moore topology, enabling communication with both adjacent and diagonal cells. A formal transition rule governs state updates for each cell based on its current state and those of its neighbors, effectively suppressing noise-induced perturbations. An additional information propagation mechanism distributes local state updates across the cellular grid, thereby improving global consistency and facilitating the progressive refinement of noisy data representations. To support robust evolutionary learning, a composite fitness function is formulated to jointly optimize classification accuracy and noise tolerance. Experimental results demonstrate that the proposed model achieves 12–18% higher accuracy and 9–14% greater noise robustness compared to standard Learning Classifier Systems variants and representative deep learning baselines. Moreover, the system exhibits approximately 20% faster convergence under high-noise conditions, while the extracted rule sets show a 15% improvement in interpretability. These findings indicate that cellular automata can serve as an effective noise-hardening mechanism within Learning Classifier Systems and provide a promising direction for developing dependable and adaptive learning algorithms in highly uncertain environments.
کلیدواژهها English