<p>In modeling tasks, one must consider not only accuracy but also timeliness. Fuzzy rough set-based feature selection aims to remove redundant features from the decision system, thereby reducing the dimensionality of the decision system and accelerating subsequent computation. However, due to data noise, the feature subset obtained by this method may retain a few unnecessary features. To address this, researchers have proposed some robust feature selection algorithms. These algorithms use wrapper techniques which are time-consuming. Inspired by the regularization method in machine learning, this paper proposes a robust feature selection algorithm with fuzzy rough set. The algorithm aims to strike an optimal balance between the roughness of the fuzzy granule and the simplicity of the model by selecting key features. It enhances or maintains classification accuracy, and requires less time during model construction. The comparative experiments demonstrate the efficiency and effectiveness of the proposed algorithm.</p>