Application of Algorithm CARDBK in Document Clustering

In the K-means clustering algorithm, each data point is uniquely placed into one category. The clustering quality is heavily dependent on the initial cluster centroid. Different initializations can yield varied results; local adjustment cannot save the clustering result from poor local optima. If there is an anomaly in a cluster, it will seriously affect the cluster mean value. The K-means clustering algorithm is only suitable for clusters with convex shapes. We therefore propose a novel clustering algorithm CARDBK—"centroid all rank distance(CARD)" which means that all centroids are sorted by distance value from one point and "BK" are the initials of "batch K-means"—in which one point not only modifies a cluster centroid nearest to this point but also modifies multiple clusters centroids adjacent to this point, and the degree of influence of a point on a cluster centroid depends on the distance value between this point and the other nearer cluster centroids. Experimental results showed that our CARDBK algorithm outperformed other algorithms when tested on a number of different data sets based on the following performance indexes: entropy, purity, F1 value, Rand index and normalized mutual information(NMI). Our algorithm manifested to be more stable, linearly scalable and faster.     

Comparing the biological coherence of network clusters identified by different detection algorithms

Protein-protein interaction networks serve to carry out basic molecular activity in the cell. Detecting the modular structures from the protein-protein interaction network is important for understanding the organization, function and dynamics of a biological system. In order to identify functional neighbor- hoods based on network topology, many network cluster identification algorithms have been devel- oped. However, each algorithm might dissect a network from a different aspect and may provide dif- ferent insight on the network partition. In order to objectively evaluate the performance of four com- monly used cluster detection algorithms: molecular complex detection (MCODE), NetworkBlast, shortest-distance clustering (SDC) and Girvan-Newman (G-N) algorithm, we compared the biological coherence of the network clusters found by these algorithms through a uniform evaluation framework. Each algorithm was utilized to find network clusters in two different protein-protein interaction net- works with various parameters. Comparison of the resulting network clusters indicates that clusters found by MCODE and SDC are of higher biological coherence than those by NetworkBlast and G-N algorithm.     

The Effective Clustering Partition Algorithm Based on the Genetic Evolution

To the problem that it is hard to determine the clustering number and the abnormal points by using the clustering validity function, an effective clustering partition model based on the genetic algorithm is built in this paper. The solution to the problem is formed by the combination of the clustering partition and the encoding samples, and the fitness function is defined by the distances among and within clusters. The clustering number and the samples in each cluster are determined and the abnormal points are distinguished by implementing the triple random crossover operator and the mutation. Based on the known sample data, the results of the novel method and the clustering validity function are compared. Numerical experiments are given and the results show that the novel method is more effective.     

ClustNails:Visual Analysis of Subspace Clusters

Subspace clustering addresses an important problem in clustering multi-dimensional data.In sparse multi-dimensional data,many dimensions are irrelevant and obscure the cluster boundaries.Subspace clustering helps by mining the clusters present in only locally relevant subsets of dimensions.However,understanding the result of subspace clustering by analysts is not trivial.In addition to the grouping information,relevant sets of dimensions and overlaps between groups,both in terms of dimensions and records,need to be analyzed.We introduce a visual subspace cluster analysis system called ClustNails.It integrates several novel visualization techniques with various user interaction facilities to support navigating and interpreting the result of subspace clustering.We demonstrate the effectiveness of the proposed system by applying it to the analysis of real world data and comparing it with existing visual subspace cluster analysis systems.     

Liquid structure of pure iron by X-ray diffraction

The liquid structure of pure iron at 1540, 1560 and 1580℃ was studied by X-ray diffraction. The results show that near the melting point there is a medium-range order structure that fades away with the increasing temperature. The average nearest distance of atoms is almost independent of the melts temperature, but the average coordination number, the atom cluster size and the atom number in an atom cluster all decrease with the increasing temperature of the melt. Near the melting point there area lot of atom clusters in the pure iron melt. The atom cluster of pure iron has the body-centered cubic lattices, which are kept from the solid state. And the body-centered cubic lattices connect into network by occupying a same edge. The atoms in the surrounding of the atom clusters are arranged disorderly.     

Clustering Categorical Data：A Cluster Ensemble Approach

Clustering categorical data, an integral part of data mining, has attracted much attention recently. In this paper, the authors formally define the categorical data clustering problem as an optimization problem from the viewpoint of cluster ensemble, and apply cluster ensemble approach for clustering categorical data. Experimental results on real datasets show that better clustering accuracy can be obtained by comparing with existing categorical data clustering algorithms.     

Using an Improved Clustering Method to Detect Anomaly Activities

In this paper, an improved k-means based clustering method （IKCM） is proposed. By refining the initial cluster centers and adjusting the number of clusters by splitting and merging procedures, it can avoid the algorithm resulting in the situation of locally optimal solution and reduce the number of clusters dependency. The IKCM has been implemented and tested. We perform experiments on KDD-99 data set. The comparison experiments with H-means＋also have been conducted. The results obtained in this study are very encouraging.     

CFSBC： Clustering in High-Dimensional Space Based on Closed Frequent Item Set

Clustering in high-dimensional space is an important domain in data mining. It is the process of discovering groups in a high-dimensional dataset, in such way, that the similarity between the elements of the same cluster is maximum and between different clusters is minimal. Many clustering algorithms are not applicable to high dimensional space for its sparseness and decline properties. Dimensionality reduction is an effective method to solve this problem. The paper proposes a novel clustering algorithm CFSBC based onclosed frequent hemsets derived from association rule mining. which can get the clustering attributes with high efficiency. The algorithm has several advantages. First, it deals effectively with the problem of dimensionality reduction. Second, it is applicable to different kinds of attributes, Third, it is suitable for very large data sets. Experiment shows that the proposed algorithm is effective and efficient     

Identification Method of Urban Road Traffic Conditions in the Internet of Vehicles Environment

In accordance with the specific deployment way of infrastructure and data exchanging technology in the Internet of vehicles(IoV),the acquiring and calculating method for three basic traffic flow parameters in IoV scenarios,including traffic flow,speed and density,was researched.Considering the complexity of traffic flow and fuzziness of human thinking,fuzzy c-means clustering algorithm based on the genetic algorithm(GA-FCM) was adopted in soft classification of urban road traffic conditions.Genetic algorithm(GA) introduced into fuzzy clustering could avoid fuzzy c-means(FCM) algorithm converging to the local infinitesimal point,which made the cluster result more precise.By means of computer simulation,data exchanging environment in IoV was imitated,and then test data set was divided into four parts.The simulation indicates that the identification method is feasible and effective for urban road traffic conditions in IoV scenarios.     

A hybrid cache placement scheme for multi-hop wireless service network

In this paper,a hybrid cache placement scheme for multihop wireless service networks is proposed. In this scheme,hot nodes in data transferring path are mined up by means of rout-ing navigation graph,and whole network is covered with network clustering scheme. A hot node has been chosen for cache place-ment in each cluster,and the nodes within a cluster access cache data with no more than two hops. The cache placement scheme reduces data access latency and workload of the server node. It also reduces the average length of data transferring,which means that fewer nodes are involved. The network system energy con-sumption decreased as involved relay nodes reduced. The per-formance analysis shows that the scheme achieves significant system performance improvement in network environment,with a large number of nodes.     

多目标监督聚类GA研究

提出了多目标监督聚类GA算法,即:根据样本的类标签有监督地将样本聚类,在每个类中根据样本属性的相似性有监督地聚成类簇.如果分属不同类标签的类簇出现相交,则相交类簇再次聚类,直到所有类簇均不相交.适应度矢量函数由类簇数和类内距离2个目标确定,类簇数和类簇中心由目标函数自动确定,从而类簇数和中心就不受主观因素的影响,并且保证了这2个关键要素的优化性质.预测分类时,删去单点类簇,并根据类簇号和离某个类簇中心距离的最近邻法则以及该类簇的类标签进行分类.算法模型采用C#实现,采用3个UCI数据集进行实例分析,实验结果表明,本算法优于著名的Native Bayes、Boost C4.5和KNN算法.     

云计算环境下关联性大数据实时流式可控聚类算法研究

针对传统聚类算法效率低、效果差和稳定性弱等弊端,提出一种新的云计算环境下关联性大数据实时流式可控聚类算法。介绍了关联性实时流式数据的定义和特点。通过粗聚类对实时抵达的数据元组进行相应的预处理,确定类簇的数量与中心点位置,形成通过存在差异的宏簇构成的集合,粗聚类采用的算法为Canopy算法。将粗聚类得到的宏簇传至K-means算法,给出了K-means算法的详细步骤,通过K-means算法完成细聚类,介绍了整个细聚类详细步骤。实验结果表明,所提算法具有效率高、质量好、稳定性强等优势,可有效实现云计算环境下关联性实时流式大数据聚类。     

一种面向不确定数据流的聚类算法

作为大数据的重要组成,产生于传感器、移动电话设备、社交网络等的不确定流数据因其具有流速可变、规模宏大、单遍扫描及不确定性等特点,传统聚类算法不能满足用户高效实时的查询要求.首先利用MBR(minimum bounding rectangle)描述不确定元组的分布特性,并提出一种基于期望距离的不确定数据流聚类算法,计算期望距离范围的上下界剪枝距离较远的簇以减少计算量;其次针对簇内元组的分布特征提出了簇MBR的概念,提出一种基于空间位置关系的聚类算法,根据不确定元组MBR和簇MBR的空间位置关系排除距离不确定元组较远的簇,从而提高聚类算法效率;最后在合成数据集和真实数据集进行实验,结果验证了所提出算法的有效性和高效性.     

一种基于自组织分级聚类的数据挖掘方法

针对传统自组织聚类方法处理数据在多维空间中存在多样性和从基本或低层次概念上发现强关联规则中的不足，提出了一种基于自组织分级聚类的数据挖掘方法．该方法采用最大似然分类自组织特征网络（MAXNET）聚类过程，利用自下而上聚合层次聚类方法，对有畸变的二值化输入模式作最大似然分类．最后对一个销售电脑商场中，与任务相关的交易数据集进行了分析，描述了从低层次概念到高层次概念的相互关系．并用matlab仿真软件把该方法和传统方法进行比较，表明了该方法的有效性．     

基于自适应邻域互信息与谱聚类的特征选择

借鉴邻域粗糙集处理连续型数据的优势,为解决传统谱聚类算法需要人工选取参数的问题,提出基于自适应邻域互信息与谱聚类的特征选择算法。首先,定义各对象在属性下的标准差集合与自适应邻域集,给出自适应邻域熵、平均邻域熵、联合熵、邻域条件熵、邻域互信息等不确定性度量,利用自适应邻域互信息对特征与标签的相关性进行排序。然后,结合共享近邻自适应谱聚类算法,将相关性强的特征聚到同一特征簇内,使不同特征簇内的特征强相异。最后,使用最小冗余最大相关技术设计特征选择算法。在10个数据集上选择特征个数与分类精度的实验结果,验证了所提算法的有效性。     

基于聚类分析和支持向量机的布匹瑕疵分类方法

提出一种基于聚类分析和支持向量机(SVM)的布匹瑕疵分类方法.该方法充分利用瑕疵的几何特征,首先使用迭代自组织数据分析技术算法(ISODATA)对其进行聚类,在聚类形成的子空间内再根据瑕疵的纹理特征利用SVM进行分类.根据布匹瑕疵的特点提出一种新的几何特征,并使用各类瑕疵的几何特征均值作为初始聚类中心,提高ISODATA算法的聚类效果.实验表明,该方法有效地提高了分类准确性,降低了训练的复杂度,分类准确率可达90%.     

一类混杂系统的模型辨识

针对一类具有切换结构的混杂系统,提出一种基于仿射传播聚类的模型辨识方法。将模型辨识问题等价成对系统数据的分类和分类数据的回归问题。通过仿射传播聚类算法对样本数据进行聚类划分,并分别采用最小二乘支持向量机算法对子样本分别建立模型。仿真结果验证了该方法的有效性和实用性。     

一种对于K-means算法的改进

数据聚类是一个功能强大的技术,它能够把数据特征相似的对象划分为一类,但是并不是所有的聚类算法的实现都能产生相同的聚类结果;并且K均值算法的结果很大程度上依赖它的初始中心的选择;提出了一种新颖的关于K均值初始中心选择的策略;该算法是基于反向最近邻(RNN)搜索,检索一个给定的数据集,其最近的邻居是一个给定的查询点中的所有点;使用这种方法计算初始聚类中心结果发现是非常接近聚类算法所需的迭代聚类中心;对提出的算法应用到K均值聚类中给予了证明;用几种流行的数据集的实验结果表明了该算法的优点。     

基于PSO的WFCM算法研究及其在医保欺诈行为发现中的应用

【目的】在没有先验知识的前提下,采用基于粒子群优化算法(PSO)的加权模糊C-均值(WFCM)聚类算法,从30多万条记录的医疗保险数据中挖掘出疑似医疗保险欺诈的记录。【方法】首先,引用改进的欧式距离、相似性函数以及交叉熵函数并通过PSO算法极小化交叉熵函数,对属性权重进行分析;其次,选取Calinski-Harabasz(CH)有效性指标,展开聚类有效性的研究;然后,基于数据预处理的结果将数据运用于PSO算法,不断更新得到各属性的权重,并运用聚类有效性评价中的CH有效性指标来动态估计最佳聚类个数,提高FCM聚类的速度;最后,将属性权重和最佳聚类数应用于FCM聚类算法,根据隶属度矩阵聚类得到疑似医疗保险欺诈结果。【结果】基于上述研究方法,本研究根据最后的隶属度矩阵来进行聚类分析。【结论】将优化的权重应用于加权FCM聚类算法与聚类有效性评价,既提高了聚类算法的高效性,又避免了主观评价对分类的影响。