<p>In the area of neighborhood rough sets, there is a challenge with the low efficiency of attribute reduction on large sample size datasets. Most existing algorithms tend to perform well only on medium and small sample size datasets. Furthermore, there has been limited research focused on identifying the smallest reduct in neighborhood rough sets. To address these issues, this paper presents the hash hypercube model, which aims to reduce the neighborhood search space for each sample, thus enhancing the efficiency of attribute reduction. Next, a random walk-based best from multiple selections mechanism is proposed to obtain a relative reduct quickly. Based on the relative reduct, a local search-based attribute reduction algorithm, combined with a dynamic attribute weighting strategy, is used to minimize the reduct continuously. To validate the effectiveness of the proposed algorithms, 16 datasets are selected from UCI for the experiments. Experiments demonstrate that our proposed attribute reduction algorithms have achieved state-of-the-art efficiency on large sample size datasets, and effectively help to find the smaller reduct with competitive classification accuracy.</p>