CAJ | 학술논문

在对数学分类模型进行分析的过程中，容易出现分类规则含义不明确，形式复杂等问题，导致传统的数学分类模型，由于采用学习算法对模型参数进行调整，无法有效实现数学分类，提出一种基于改进模糊粗糙集的数学分类模型，在模糊信息观下对模糊粗糙集进行分析。将互信息引入模糊粗糙集的分析中，对模糊决策表中模糊属性的相对重要性进行度量。通过bottom-up形式对相对约简进行计算。将空集作为初始点，依据属性重要性，逐次获取重要属性将其添加至集合中，直至达到终止条件。通过DTRS对数据集进行属性约简，将约简结果当成模型的输入变量。对数学分类模型的输入变量进行离散化处理。对决策表中的重复样本及通过冗余条件获取的决策表进行删除，获取决策规则。通过设定阈值对置信度较低的模糊规则进行过滤，删除因噪声样本形成的错误规则。仿真实验结果表明，所提方法具有很高的分类精度。
재대수학분류모형진행분석적과정중，용역출현분류규칙함의불명학，형식복잡등문제，도치전통적수학분류모형，유우채용학습산법대모형삼수진행조정，무법유효실현수학분류，제출일충기우개진모호조조집적수학분류모형，재모호신식관하대모호조조집진행분석。장호신식인입모호조조집적분석중，대모호결책표중모호속성적상대중요성진행도량。통과bottom-up형식대상대약간진행계산。장공집작위초시점，의거속성중요성，축차획취중요속성장기첨가지집합중，직지체도종지조건。통과DTRS대수거집진행속성약간，장약간결과당성모형적수입변량。대수학분류모형적수입변량진행리산화처리。대결책표중적중복양본급통과용여조건획취적결책표진행산제，획취결책규칙。통과설정역치대치신도교저적모호규칙진행과려，산제인조성양본형성적착오규칙。방진실험결과표명，소제방법구유흔고적분류정도。
In the process of analysis of the classification model, it's easy to have a classification rules meaning is not clear, the complex form and so on, led to the traditional classification of mathematical model, as a result of learning algorithm to adjust model parameters, unable to effectively implement mathematical classification, in this paper, a classification model based on improved fuzzy rough sets, under the concept of fuzzy information analysis of fuzzy rough sets.The mutual information is introduced into the fuzzy rough set analysis, fuzzy attributes in fuzzy decision table the relative importance of measurement.Through the bottom- up form of relative reduction is calculated.Will be an empty set as the initial point, on the basis of attribute importance, successive important attributes, add it to the collection until termination condition is reached.Through the DTRS to attribute reduction of data set, the reduction results as an input variable of the model.Classification of mathematics model of the input variables discretization processing.In decision table by means of duplicate samples and redundant get deleted, decision table to get decision rules.By setting the threshold value of confidence lower fuzzy rules to filter and remove noise samples form error rule.The simulation results show that the proposed method is of high classification accuracy.