Abstract: In the practical application of the multi-category classification problem, when the cognition of the decision object changes from coarse-grained to fine-grained, by using the granular structure, a sequential three-way decisions model based on multi-category classification is proposed. On this basis, it is time-consuming to computer sequential three-way decisions with the non-incremental method of this model. For the change of conditional attributes in the decision table, the incremental method of the model is given to make the decision. First, the condition attribute of each level is obtained by adding new attributes, and a multilevel granular structure is constructed. Under the multilevel granular structure, the loss function matrix of each level decision table is given. Then thresholds of each decision class are calculated and decisions of each level decision table are made in order. Finally, a sequential three-way decisions algorithm for multi-category classification is given. An example is used to illustrate the calculation process of the   algorithm. In order to dynamically update a sequence of three regions, an incremental algorithm for calculating      sequential three-way decisions is given based on multi-category classifications. The effectiveness of the proposed method is verified by simulation experiments.

Key words: sequential three-way decision, granular structure, multi-category classification, incremental approach

摘要： 多类分类问题的实际应用中，在决策对象的认识由粗粒度向细粒度转化时，通过使用粒结构，提出一种基于多类分类的序贯三支决策模型。在此基础上，使用该模型非增量的方法计算序贯三支决策的时间开销较大，针对决策表中条件属性的变化，给出该模型的增量方法来进行决策。首先，通过增加新属性得到每层的条件属性，构建多层次粒结构。在多层次粒结构下，给出每层决策表的损失函数矩阵。然后，按层依次计算决策表中每个决策类的阈值，进行决策表的三支决策。最后，给出多类分类的序贯三支决策算法，通过实例说明该算法的计算过程。为了动态更新多层次下的三个域集，基于多类分类给出计算序贯三支决策的增量算法，通过仿真实验验证了该方法的有效性。

关键词: 序贯三支决策, 粒结构, 多类分类, 增量方法