<p>Feature selection in high-dimensional data is a persistent challenge, as the curse of dimensionality often renders traditional methods ineffective. The performance of neighborhood rough set (NRS) is limited in high-dimensional feature selection as traditional distance metrics become ineffective. To address this, this paper proposes a synergistic optimization framework based on the firefly algorithm (FA), termed NRSOFA, designed to solve the challenge of high-dimensional adaptability of NRS. The framework synergistically integrates three key mechanisms: first, a normalized distance metric is adopted to overcome the effects of the curse of dimensionality; second, an NRS dependency-guided population optimization strategy is introduced to improve solution quality; and finally, a local reverse cooperative evolution mechanism effectively escapes local optima. Comprehensive experiments on 15 standard datasets demonstrate that NRSOFA shows significant advantages in both the accuracy and stability of feature selection compared against a diverse suite of established baselines, including classical neighborhood rough set models, adaptive neighborhood rough set variants, and hybrid rough set/swarm intelligence approaches. The study confirms that through multi-strategy synergy, the framework successfully overcomes the application bottleneck of traditional NRS in high-dimensional spaces, providing a robust and efficient solution for complex feature selection problems.</p>