面向不确定性数据分析的三支决策粗糙集模型、算法和应用研究

As an efficient solution for uncertainty representation, learning, and reasoning, the three-way decision-theoretic rough set theory has drawn a lot of attentions and obtained some progress. However, there are still some problems in this area to be solved, such as the uncertainty measures are studied independently, the model construction method is too simple, and the evaluation measures for the model are not complete. To address these problems, this project intends to carry out theoretical research from the following three aspects: (1) Studying the definitions of different uncertainty measures and analyzing their relationships; (2) Constructing a generalized three-way decisions based classification model based on given uncertainty measures; (3) Designing the three-way decisions converting algorithms for reducing uncertainty and completing the evaluation measure system. At the same time, based on the proposed three-way decision-theoretic rough set model, this project will take the brain neuroimaging data to study the prediction of consciousness state and recovery outcome for acquired brain injury subjects, which can give a strong decision support for scientific and rational medical diagnosis. This project has important scientific significance and practical application value to promote the research of analyzing uncertainty in the artificial intelligence area.

三支决策粗糙集理论是不确定性知识表示、学习和推理的一种有效解决方案，已有研究虽然取得了很多进展，但仍存在不确定性度量不成体系、模型构建方式过于单一、评价指标体系不够完备等问题。针对于此，本项目拟从以下三个方面展开理论研究：（1）研究三支决策粗糙集模型下不确定性度量的表示及其关联性；（2）构建基于不确定性度量的通用三支决策分类模型；（3）设计面向减小不确定性的三支决策转化算法及评价指标体系。在理论研究的同时，本项目将以大脑神经影像数据为研究对象，基于所提三支决策粗糙集模型，对脑损伤病人意识状态评判和苏醒预测等问题开展应用研究，为科学合理的医疗诊断提供决策支持。本项目对推动人工智能领域不确定性分析研究具有重要的科学意义和实际应用价值。

三支决策粗糙集理论是不确定性知识表示、学习和推理的一种有效解决方案，已有研究虽然取得了很多进展，但仍存在不确定性度量不成体系、模型构建方式过于单一等问题。针对于此，本项目从以下三个方面展开研究：（1）研究三支决策粗糙集模型下不确定性度量的表示及其关联性；（2）构建基于不确定性度量的三支决策分类模型及阈值求解；（3）设计面向减小不确定性的三支决策转化算法及应用研究。基于本项目研究，提出了一系列用于刻画区间集值信息表及直觉模糊粗糙集不确定性的度量，设计了一系列基于不确定性度量的属性约简算法、分类算法和聚类算法。此外，在对三支决策粗糙集理论研究的同时，本项目对标记多义性带来的标记不确定性问题也展开了研究，以标记分布学习理论为研究对象，提出了一系列基于标记相关性的标记分布学习算法和标记增强算法。在应用方面，以大脑神经影像数据为研究对象，对自闭症辅助诊断等问题开展应用研究，为科学合理的医疗诊断提供决策支持。本项目对推动人工智能领域不确定性分析研究具有重要的科学意义和实际应用价值。