Kernel Parameter Optimization for Semi-Supervised Fuzzy Clustering with Pairwise Constraints

A critical problem related to semisupervised kernel clustering is the selection of an optimal kernel parameter since the value of parameter has significant impact on the performance of clustering. In this paper, we construct a semi-supervised kernel fuzzy c-means clustering algorithm in terms of pairwise constraints to obtain an optimal kernel parameter. Combined with kernel parameter initialization directly using the given constraints, a new optimization process is derived to automatically estimate the optimal parameter of kernel function. Experimental results show that, with the effective use of pairwise constraints, the proposed approach works well for the estimation of kernel parameter in semi-supervised kernel fuzzy c-means clustering.     

New method of mining incomplete data

The data used in the process of knowledge discovery often includes noise and incomplete information. The boundaries of different classes of these data are blur and unobvious. When these data are clustered or classified, we often get the coverings instead of the partitions, and it usually makes our information system insecure. In this paper, optimal partitioning of incomplete data is researched. Firstly, the relationship of set cover and set partition is discussed, and the distance between set cover and set partition is defined. Secondly, the optimal partitioning of given cover is researched by the combing and parting method, acquiring the optimal partition from three different partitions set family is discussed. Finally, the corresponding optimal algorithm is given. The real wireless signals offten contain a lot of noise, and there are many errors in boundaries when these data is clustered based on the tradional method. In our experimant, the proposed method improves correct rate greatly, and the experimental results demonstrate the method’s validity.     

A Fuzzy Similarity-based Clustering Optimized by Particle Swarm Optimization

Traditional fuzzy clustering algorithms based on objective function is unable to determine the optimum number of clusters, sensitive to the initial cluster centers, and easily sunk into the issue of local optimum. A Fuzzy similarity-based clustering （FSBC） algorithm is proposed in this paper. This method consists three phases： first, the objective function is modified by integrating Fuzzy C-means （FCM） and Possibilistic C-means （PCM） method; second, using the density function from data for similarity-based clustering to automatically generate initial prototype without requesting users to specify; finally, the iteration process optimized by Particle swarm optimization （PSO） to obtain appropriate adjustment parameters that can provide better results, which avoids the local minimum problems of traditional methods. The experimental results on the synthetic data and UCI standard data sets show that the proposed algorithm has greater searching capability, less computational complexity, higher clustering precision.     

Clustering of DOA data in radar pulse based on SOFM and CDbw

Clustering is the main method of deinterleaving of radar pulse using multi-parameter. However, the problem in clustering of radar pulses lies in finding the right number of clusters. To solve this problem, a method is proposed based on Self-Organizing Feature Maps （SOFM） and Composed Density between and within clusters （CDbw）. This method firstly extracts the feature of Direction Of Arrival （DOA） data by SOFM using the characteristic of DOA parameter, and then cluster of SOFM. Through computing the cluster validity index CDbw, the right number of clusters is found. The results of simulation show that the method is effective in sorting the data of DOA.     

The optimal time-frequency atom search based on a modified ant colony algorithm

this paper, a new optimal time-frequency atom search method based on a modified ant colony algorithm is proposed to improve the precision of the traditional methods. First, the discretization formula of finite length time-frequency atom is inferred at length. Second; a modified ant colony algorithm in continuous space is proposed. Finally, the optimal time- frequency atom search algorithm based on the modified ant colony algorithm is described in detail and the simulation experiment is carried on. The result indicates that the developed algorithm is valid and stable, and the precision of the method is higher than that of the traditional method.     

Research on Loophole with Second Distribution of Real Value Detectors

Traditional anomaly detection algorithm has improved to some degree the mechanism of negative selection.There still remain many problems such as the randomness of detector generation,incompleteness of selfset and the generalization ability of detectors,which would cause a lot of loopholes in non-self space.A heuristic algorithm based on the second distribution of real value detectors for the remains of loopholes of the non-self space in the first distribution and the mutation regions of self space is proposed.The algorithm can distribute real value detectors through omission data based on the methods of partition and movement.A method is proposed to solve the problem on how to get the optimal solutions to the parameters related in the algorithm.Theoretical analysis and experimental results prove the universality and effectiveness of the method.It is found that our algorithm can effectively avoid the generation of loopholes and thus reduce the omission rate of detector sets.     

Entropy-like distance driven fuzzy clustering with local information constraints for image segmentation

To improve the anti-noise ability of fuzzy local information C-means clustering, a robust entropy-like distance driven fuzzy clustering with local information is proposed. This paper firstly uses Jensen-Shannon divergence to induce a symmetric entropy-like divergence. Then the root of entropy-like divergence is proved to be a distance measure, and it is applied to existing fuzzy C-means (FCM) clustering to obtain a new entropy-like divergence driven fuzzy clustering, meanwhile its convergence is strictly proved by Zangwill theorem. In the end, a robust fuzzy clustering by combing local information with entropy-like distance is constructed to segment image with noise. Experimental results show that the proposed algorithm has better segmentation accuracy and robustness against noise than existing state-of-the-art fuzzy clustering-related segmentation algorithm in the presence of noise.     

CONSIDERING NEIGHBORHOOD INFORMATION IN IMAGE FUZZY CLUSTERING

Fuzzy C-means clustering algorithm is a classical non-supervised classification method. For image classification, fuzzy C-means clustering algorithm makes decisions on a pixel-by-pixel basis and does not take advantage of spatial information, regardless of the pixels' correlation. In this letter, a novel fuzzy C-means clustering algorithm is introduced, which is based on image's neighborhood system. During classification procedure, the novel algorithm regards all pixels' fuzzy membership as a random field. The neighboring pixels' fuzzy membership information is used for the algorithm's iteration procedure. As a result, the algorithm gives a more smooth classification result and cuts down the computation time.     

A New Word Clustering Algorithm Based on Word Similarity

Category-based statistic language model is an important method to solve the problem of sparse data in statistical language models. But there are two bottlenecks about this model: 1) The problem of word clustering, it is hard to find a suitable clustering method that has good performance and has not large amount of computation; 2) Class-based method always loses some prediction ability to adapt the text of different domain. In order to solve above problems, a novel definition of word similarity by utilizing mutual information was presented. Based on word similarity, the definition of word set similarity was given and a bottom-up hierarchical clustering algorithm was proposed. Experimental results show that the word clustering algorithm based on word similarity is better than conventional greedy clustering method in speed and performance, the perplexity is reduced from 283 to 207.8.     

FUZZY PARTITIONING OF FEATURE SPACE FOR PATTERN CLASSIFICATION BASED ON SUPERVISED C1USTERING

The choice of a fuzzy partitioning is crucial to the performance of a fuzzy system based on if-then rules. However, most of the existing methods are complicated or lead ,o too many subspaces, which is unfit for the applications of pattern classification. A simple but effective clustering approach is proposed in this paper, which obtains a set of compact subspaces and is applicable for classification problems with higher dimensional feature. Its effectiveness is demonstrated by the experimental results.     

模糊c-均值聚类算法中加权指数m的研究

 加权指数m是模糊c-均值(FCM)聚类算法中的一个重要参数.本文从FCM算法出发研究了m对聚类分析的影响,m的最佳选取方法及其在聚类有效性中的应用三个问题.实验结果表明:m不合适的取值将严重影响算法的性能;在实际应用中m的最佳取值范围为 ,这与Pal的实验结论相一致;另外基于最优加权指数m^*的类别数确定方法是相当有效和灵敏的.     

基于最优分类系数及分类熵准则的模糊C均值聚类算法

本文分别以模糊分类系数和分类熵为聚类有效性的测度函数，采用模糊Ｃ均值分类法，求最优的测度函数值所代表的全局极值的那种分类，确定为最佳分类方，对一张由卫星拍摄的地面ＭＩＧ－２９型飞机照片进行了成功的分类试验。     

关于FCM算法中的权重指数m的一点注记

模糊c均值算法(FCM)是经常使用的聚类算法之一.使用模糊c均值算法时,如何选取模糊指标m一直是一个悬而未决的问题.部分文献根据实验结果建议最佳的权重指数可能位于区间 ,但大多数研究者使用m=2.本文阐述了FCM算法有效性与聚类有效性之间的理论联系,指出如果某个权重指数使得FCM算法作为聚类算法不能有效工作,则其不能作为最佳的权重指数.据此,我们进行了数据实验,数据实验结果说明了权重指数的最佳取值未必位于区间 .     

基于粒子群优化非局部模糊聚类图像分割算法

为提高局部模糊聚类算法(WFLICM)对噪声图像 分割的抗噪性,克服模糊聚类图像分割算法对初 始聚类中心的敏感性及易陷入局部最优问题,在WFLICM算法的基础上提出一种基于粒子群 优化的融合 局部和非局部空间信息的模糊聚类图像分割算法(PSO-WMNLFCM)。首先,利用粒子群优化 算法的全局 寻优能力得到最优粒子,并以此粒子作为模糊聚类算法的初始聚类中心。其次,用像素的非 局部空间信息 替换模糊因子中的局部邻域值,产生新的目标函数。最后,由拉格朗日乘子法最小化目标函 数,得到隶属 度和聚类中心的更新公式,从而完成图像分割。仿真结果表明,PSO-WMNLFCM算法相比于 模糊局部聚 类(FLICM)算法、局部模糊权重(WFLICM)算法、非局部模糊聚类(NLFCM)算法、非局部模 糊聚类 (MNLFCM)算法、基于粒子 群的局部模糊聚类(PSO-FLICM)算法的划分系数提高了20.92%,20.51%,24.84%,1.44%,23.28%左右。     

改进的模糊核C-均值算法

将核方法的思想推广到模糊C-均值算法，提出一种改进的模糊核C-均值算法。改进的模糊核C-均值算法较以前的模糊核C-均值方法有更好的鲁棒性，不但可以在有野值存在的情况下得到较好的聚类结果．而且因为放松的隶属度条件，使最终聚类结果对预先确定的聚类数目不十分敏感。改进的模糊核C-均值算法在多种数据结构条件下可以有效地进行聚类。     

基于类内差异和类间重叠的有效性函数

模糊C-均值(FCM)聚类算法的一个主要问题是需要事先确定聚类的数目,为此定义了类内差异度和类间重叠度来分别度量同一个聚类中数据的相似度和不同聚类间的分离程度,进而基于这两个度量提出一个新的有效性函数用于判定最佳聚类数目。实验结果表明,该有效性函数能有效地判定聚类数目,并且有较好的鲁棒性。     

一种基于属性空间相似性的模糊聚类算法

模糊C均值(FCM)聚类算法及其相关改进算法基于最大模糊隶属度原则确定聚类结果,没有充分利用迭代后的模糊隶属度矩阵和簇类中心的样本属性特征信息,影响聚类准确度。针对这个问题,该文提出一种新的改进思路:改进FCM算法输出定类原则。给出二元属性拓扑子空间中属性相似度的定义,最终提出一种基于属性空间相似性的改进FCM算法(FCM-SAS):首先,选择FCM算法聚类后模糊隶属度低于聚类置信度的样本作为存疑样本;然后,计算存疑样本与聚类后聚类中心的属性相似度;最后,基于最大属性相似度原则更新存疑样本的簇类标签。通过UCI数据集实验,证明算法不仅有效,还较一些基于最大模糊隶属度原则定类的改进算法具有更优的聚类评价指标。     

一种基于调和均值的模糊聚类算法

k调和均值算法用数据点与所有聚类中心的距离的调和平均替代了数据点与聚类中心的最小距离，是一种减小初始值影响聚类结果的有效的聚类方法。本文对k调和均值算法进行扩展，考虑到数据点同时对不同聚类的隶属关系，将模糊的概念应用到聚类中，提出了模糊k调和均值-Fuzzv K—Harmonic Means(FKHM)算法。在中心迭代聚类算法的统一框架的基础上，推导出FKHM算法聚类中心的条件概率表达式以及在迭代过程中的数据点加权函数表达式。以划分相似度作为聚类结果的评价准则，实验表明，FKHM算法在聚类对于初值不敏感的同时提高了聚类结果的精确度，达到较好的聚类效果。     

一种启发式变阶直觉模糊时间序列预测模型

论文针对已有高阶模糊时间序列模型在预测精度和预测范围上的限制,结合直觉模糊集理论,提出一种启发式变阶直觉模糊时间序列预测模型。模型首先应用直接模糊聚类算法对论域进行非等分划分;然后,针对直觉模糊时间序列的数据特性,改进现有直觉模糊集隶属度和非隶属度函数的建立方法;最后,采用阶数随序列实时变化的高阶预测规则进行预测,并将历史数据发展趋势的启发知识引入解模糊过程,使模型的预测范围得到扩展。在Alabama大学入学人数和北京市日均气温两组数据集上分别与典型方法进行对比实验,结果表明该模型有效克服了传统模型的缺点,拥有较高的预测精度,证明了模型的有效性和优越性。     

用自动模糊划分与改进Apriori算法生成QAR关联规则

针对属性粒度模糊划分需事先给定与Aprirori算法效率低的问题,提出基于自动模糊划分和改进Apriori算法的QAR关联规则生成方法。首先对QAR数据进行空缺值填补等预处理;然后给出最佳聚类准则并根据给出的最佳聚类准则得到最佳聚类,从而对QAR属性完成自动模糊划分及隶属函数的确定;之后通过记录数据项位置及简化连接与剪枝过程来提高Apriori算法的效率;并将其应用到QAR关联规则的生成过程;最后通过品质和性能度量两方面的实验,表明此方法在各方面的性能均优于经典方法。