<p>In recent years, hierarchical classification has become one of the research hotspots in the field of artificial intelligence, as it effectively integrates the hierarchical structural information between data categories to improve classification rationality. However, in practical applications, the curse of dimensionality leads to the degradation of classification performance, making feature selection a core and critical step for achieving effective classification. Most existing hierarchical feature selection methods focus on the multi-granularity characteristics of decision attributes, but they ignore the defect that single-granularity conditional attributes are difficult to adapt to the differentiated division of objects under different decision granularities, which ultimately results in knowledge representation bias and increased model uncertainty. Therefore, this paper proposes a hierarchical classification feature selection strategy based on multi-granularity concepts (MGCHCFS). First, according to the high-dimensional characteristics of hierarchical data, an aggregated granulation method is adopted for attributes by utilizing their correlation information, thereby constructing a multi-granularity regular fuzzy-crisp formal decision context. Then, the overall hierarchical classification task is decomposed into multiple subtasks. In each subtask, a multi-granularity concept space is generated based on the constructed multi-granularity regular fuzzy-crisp formal decision context, and the conditional information entropy is calculated in this space to select the optimal granularity. Finally, experimental verification demonstrates that the proposed MGCHCFS strategy can effectively improve the performance of hierarchical classification tasks.</p>