Attribute reduction, as a technique for selecting qualified attributes which can satisfy the intended constraint related to considered measure, has been widely explored. Notably, one and only one reduct is derived through using one searching strategy in most cases. Nevertheless, only one reduct may be not enough for us to evaluate its effectiveness. To fill such gap, an approach of crosswise computing reduct is proposed for obtaining multiple reducts. The computation of reduct is realized through partitioning the whole data into several groups, and crosswise selecting some groups to form different subsets of data, then computing reducts over these different subsets of data. Moreover, to speed up the process of crosswise computing reduct, an acceleration strategy is designed. The main thinking of our acceleration strategy is to compute the reduct over different subsets of data on the basis of reduct over the whole data. The experimental results over 16 data sets show the following superiorities of our strategy: (1) our approach can decrease the elapsed time of crosswise computing reducts significantly; (2) our approach can not only provide reduct with higher stability, but also maintain the classification performance; (3) the attributes in reduct can provide more stable classification results.

作为一种选择能够满足与所考虑的度量有关的预期约束的合格属性的技术，属性缩减已被广泛探索。值得注意的是，在大多数情况下，通过使用一种搜索策略可以得出一种且只有一种还原。然而，仅减少一项措施可能不足以使我们评估其有效性。为了填补这一空白，提出了一种横向计算约简的方法来获得多个约简。归约计算是通过将整个数据划分为几个组，然后交叉选择一些组以形成不同的数据子集，然后对这些不同的数据子集进行归约计算而实现的。此外，为了加快横向计算约简的过程，设计了一种加速策略。我们的加速策略的主要思想是在对整个数据进行归约的基础上，计算不同数据子集的归约。在16个数据集上的实验结果表明，我们的策略具有以下优势：（1）我们的方法可以大大减少横向计算约简的时间。（2）我们的方法不仅可以为还原提供更高的稳定性，而且可以保持分类性能；（3）归约中的属性可以提供更稳定的分类结果。而且还能保持分类性能；（3）归约中的属性可以提供更稳定的分类结果。而且还能保持分类性能；（3）归约中的属性可以提供更稳定的分类结果。