Abstract:

To compensate for the deficiency of multi-granularity decision rough sets in dealing with uncertain information, this paper proposes a multi-granulation decision-theoretic rough set analysis method based on supervised mechanism. This method introduces intra-class and inter-class in multi-granularity decision rough sets in view of the advantages of supervised learning, that is, being able to involve the existing or predicted category label information of the object. The lower and upper approximations of the model are provided. Relevant properties and conclusions are proved. The effectiveness and reliability of the method are verified by the case of construction site project construction. The experimental results show that by adjusting two kinds of thresholds, the fault tolerance and classification ability of the revised version can be further improved.

Key words: supervisory mechanism, multi-granulation decision-theoretic rough set, probability rough set

摘要：

为弥补多粒度决策粗糙集刻画不确定性知识能力的不足，鉴于监督学习能够考虑对象现有或预测的类别标签信息，在多粒度决策粗糙集中引入类内阈值和类间阈值的概念，提出了基于监督机制的多粒度决策粗糙集，给出模型的下、上近似，并对相关性质和结论进行证明。以工地项目建设的实例验证了模型的有效性与可靠性。实验结果表明，通过调整类内阈值和类间阈值，可进一步提高原模型的容错和分类能力。

关键词: 监督机制, 多粒度决策粗糙集, 概率粗糙集