基于模糊粗糙集的复杂数据规则提取方法及应用研究

Rule acquisition is one of the most important techniques for discovering knowledge in complex data. Granular computing based on rough sets provides a solid theoretical foundation and some effective methods for rule acquisition. In order to extract granular rules from a complex data for decision making, we base the fuzzy rough sets to make a series of studies on granular rule acquisition and its applications. Specifically, three aspects need to be investigated: The granular structures of fuzzy rough sets are employed to construct the fuzzy granular rules. A fuzzy granular rule acquisition oriented attribute reduction framework for fuzzy decision systems is formulated, and the corresponding attribute reduction algorithm is provided; The similarity between the fuzzy granular rules is measured, and a strategy of merging similar rules is provided. Based on the merged rules, a rule-based classifier is designed, and the generalization capability of the classifier is assessed; Under dynamic environment, via the grasp of the variation of the antecedents of fuzzy granular rules, an incremental fuzzy granular rule acquisition algorithm is presented and its effectiveness is evaluated. In conclusion, the main objective of this project is to provide, from the viewpoint of rule acquisition, some feasible and effective rough sets-based methods of knowledge discovering and decision making for complex data.

规则提取是复杂数据知识发现的一种重要手段，而基于粗糙集的粒计算为规则提取的研究提供了坚实的理论依据和有效的方法。本项目将基于模糊粗糙集对复杂数据进行粒规则的提取并围绕提取的规则在应用方面展开研究，具体内容包括：基于模糊粗糙集的粒结构构造模糊粒规则，建立面向模糊粒规则提取的复杂数据属性约简框架并给出相应算法；度量模糊粒规则间的相似性并给出合并相似规则的策略，基于得到的精简规则集设计一个规则分类器，并对其泛化能力进行评估；在动态环境下，通过对模糊粒规则前件的变化规律的掌握，提出增量式模糊粒规则提取的算法并对其有效性进行评估。综上，本项目的研究将针对复杂数据，从规则提取的角度提出一些可行且有效的知识发现及分类决策的粗糙集方法。

模糊粗糙集理论作为经典粗糙集理论的一种有用推广可以用来处理实数值甚至是混合值的复杂数据，该理论已在机器学习和数据挖掘领域得到了广泛的应用。本项目基于模糊粗糙集主要在模糊粒规则的提取和应用、快速特征选择、增量特征选择等知识发现方面展开了研究：（1）利用模糊下近似的粒结构构建了模糊粒规则，进一步通过模糊粒规则的覆盖能力针对整个数据集筛选代表样例，实现数据压缩，并提出了面向代表样例的模糊粗糙集特征选择方法，该方法针对较大规模数据集可以快速地选择出关键特征；（2）分别从模糊粗糙集信息熵的加速计算和利用模糊下近似值对样例进行筛选的角度给出了基于模糊粗糙集信息熵的快速特征选择算法，从而节约了特征选择的计算时间；（3）针对动态新增的样例集，利用模糊粗糙集产生的模糊粒的信息覆盖的特性，给出了一种从新增样例中主动筛选代表样例的方法，并研究了模糊粗糙集信息熵的增量机制，进而给出了一种基于模糊粗糙集的主动增量特征选择方法，该方法对含有大量样例的数据集有明显的计算时间方面的优势。. 本课题基本完成了各项研究内容，达到了预期的研究目标。截至目前，已在国际国内重要学术期刊上发表或接受论文8篇，其中SCI论文6篇。整体而言，本项目的研究主要基于模糊粗糙集为决策分析者在模糊粒规则的提取和应用、快速特征选择、主动增量特征选择等方面提供了一些可行有效的知识发现方法，拓展了模糊粗糙集的实际应用。.