基于不一致邻域的批增量式属性约简

摘要

经典粗糙集仅能处理离散性数据,而邻域粗糙集通过运用距离函数解决了此局限性.基于此,提出一种信息观下批增量式属性约简算法.分析批量增加样本后新样本集下条件熵的变化机制,给出条件熵的计算公式;通过公式得出新加入样本的不一致邻域引起条件熵的变化,当新增样本加入到原样本集后,只需找到新增样本集的不一致邻域,与新增样本集一起进行约简,对原约简集进行冗余剔除,得到最终约简.该算法避免了有重复的约简,大大减少了计算量.将该算法应用到UCI数据集以及某单位的科技人才流动绩效评价指标体系中进行实验,实验结果验证了该算法的有效性和高效性.%Neighborhood rough set uses distance function systematically to solve a limitation that the classical rough set can only deal with discrete data.Based on this,a batch of incremental attribute reduction algorithm from the information view was presented.The changing rules of the conditional entropy after adding the batch of new samples were analyzed,and the calculation formula of conditional entropy was given out.The result that the changes of condition entropy are caused by inconsistent neighborhood of a batch of new samples was concluded.When the new samples were added to the original set,the only thing to do was to find inconsistent neighborhood of new samples and reduce it with the new sample set,the original reduction set was reduced and the final reduction was got,avoiding repeating reduction and saving massive amount of calculation.It is verified to be very effective and efficient by the application with this algorithm both in UCI data set and an evaluation system for scientific talents' performance.