Abstract:

Aiming at generalization of attribute reducts, the class-specific attribute reduct and its heuristic algorithm are established by neighborhood approximation condition-entropy. High classification-based neighborhood approximation condition-entropy is decomposed, and middle class-specific neighborhood approximation condition-entropy is extracted to achieve its double bounds and granulation non-monotonicity. Based on the new information measure, the class-specific attribute reduct is proposed, and a heuristic algorithm is accordingly designed. The relevant validity is finally verified by decision table examples and data set experiments. The obtained results are useful for uncertainty measurement and optimization application of class-specific pattern recognition.

Key words: attribute reduction, class-specific attribute reduction, heuristic reduction algorithm, neighborhood rough set, neighborhood approximation condition-entropy, granular computing

摘要：

针对属性约简的推广，采用邻域近似条件熵，建立邻域粗糙集的特定类属性约简及其启发算法。粒化分解决策分类的高层邻域近似条件熵，提取定义特定类的中层邻域近似条件熵并得到上下界与粒化非单调性；提出基于邻域近似条件熵的特定类属性约简，设计启发式约简算法；采用决策表实例与数据集实验进行有效验证。所得结果有利于特定类模式识别的不确定性度量与优化应用。

关键词: 属性约简, 特定类属性约简, 启发式约简算法, 邻域粗糙集, 邻域近似条件熵, 粒计算