用于不平衡数据分类的0阶TSK型模糊系统

发布时间：2018-02-11 02:13

本文关键词： 不平衡数据分类马尔科夫蒙特卡洛 Takagi-Sugeno-Kang型模糊系统　出处：《自动化学报》2017年10期 　论文类型：期刊论文

【摘要】：处理不平衡数据分类时,传统模糊系统对少数类样本识别率较低.针对这一问题,首先,在前件参数学习上,提出了竞争贝叶斯模糊聚类(Bayesian fuzzy clustering based on competitive learning,BFCCL)算法,BFCCL算法考虑不同类别样本聚类中心间的排斥作用,采用交替迭代的执行方式并通过马尔科夫蒙特卡洛方法获得模型参数最优解.其次,在后件参数学习上,基于大间隔的策略并通过参数调节使得少数类到分类面的距离大于多数类到分类面的距离,该方法能有效纠正分类面的偏移.基于上述思想以0阶TSK型模糊系统为具体研究对象构造了适用于不平衡数据分类问题的0阶TSK型模糊系统(0-TSK-IDC).人工和真实医学数据集实验结果表明,0-TSK-IDC在不平衡数据分类问题中对少数类和多数类均具有较高的识别率,且具有良好的鲁棒性和可解释性.
[Abstract]:When dealing with unbalanced data classification, the traditional fuzzy system has a low recognition rate for a few samples. In order to solve this problem, first of all, in the former part parameter learning, In this paper, the Bayesian fuzzy clustering based on competitive learning BFCCL (Bayesian fuzzy clustering based on competitive learning) algorithm is proposed, which takes into account the repulsive effect between different clustering centers. The optimal solution of model parameters is obtained by means of alternating iteration and Markov Monte Carlo method. Secondly, in the latter part parameter learning, Based on the large interval strategy and parameter adjustment, the distance between a few classes and the classification surface is larger than that between the majority class and the classification surface. This method can effectively correct the deviation of classification surface. Based on the above idea, a 0-order TSK fuzzy system for unbalanced data classification is constructed based on the above idea. The artificial and real medical numbers are obtained. According to the set experiment results, it is shown that the 0-TSK-IDC has a high recognition rate for a few classes and a large number of classes in the problem of unbalanced data classification. And it has good robustness and interpretability.
【作者单位】：江南大学数字媒体学院;常州大学信息科学与工程学院;
【基金】：国家自然科学基金(61502058,61572085,61572236) 江苏省自然科学基金资助(BK20160187) 中央高校基本科研业务费专项资金资助项目(JUSRP51614A)资助~~
【分类号】：TP311.13

【相似文献】