Neighborhood rough sets are proficient in managing continuous data, but the selection of the neighborhood radius often necessitates multiple tests, leading to substantial time expenditure. To solve this problem, this paper proposes an acceleration algorithm for class-specific attribute reduction. Initially, the concept of class-specific multi-granularity reduction within the framework of neighborhood rough sets is clarified. The relationship between reductions induced by different neighborhood radii is demonstrated by analyzing the variation of approximate quality across different granularities. Subsequently, by integrating local neighborhood rough sets with sequential three-way decision, the sets of candidate samples and attributes are simultaneously reduced at a single granularity, thereby further diminishing the computational complexity. Finally, the corresponding multi-granularity acceleration algorithm is proposed. The experimental results derived from ten UCI datasets indicate that the proposed algorithm significantly decreases the time required for class-specific attribute reduction while maintaining the classification capability of neighborhood decision systems.