Search

Medical Data Mining is a very active and challenging research area in Data Mining community. However researchers entering Medical Data Mining should be aware that in core clinical, dentistry and nursing, data mining is not welcomed as much as we believe and publication of results in these journals based on Data Mining algorithms is not easily possible. In this paper, in addition to presenting one of our "successful" KDD projects in Urology that did not get to anywhere, we back up our belief based on designed searches on PubMed and review literature based on these searches. Our findings suggest that few Data Mining algorithms made their ways into core clinical journals. The paper concludes by reasons we have collected through our experiences. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a new type of reducts called the λ-Fuzzy-Reduct, where the fuzzy similarity relation is constructed by means of cosine-distances of decision vectors and the parameter λ is used to tune the similarity precision level. The λ-Fuzzy-Reduct can eliminate harsh requirements of the distribution reduct, and it is more flexible than the maximum distribution reduct, the traditional reduct, and the generalized decision reduct. Furthermore, we prove that the distribution reduct, the maximum distribution reduct, and the generalized decision reduct can be converted into the traditional reduct. Thus in practice the implementations of knowledge reductions for the three types of reducts can be unified into efficient heuristic algorithms for the traditional reduct. We illustrate concepts and methods proposed in this paper by an example. [ABSTRACT FROM AUTHOR]

In this paper, a music information retrieval system in real mobile environment is proposed. In order to alleviate distortions due to the mobile noise, a noise reduction algorithm is applied and then a feature extraction using Multi-Feature Clustering is implemented to improve the system performance. The proposed system shows quite successful performance with real world cellular phone data. [ABSTRACT FROM AUTHOR]

Incrementally maintaining association rules based on two or more classes of frequent item sets may reduce the costs of scanning the original database remarkably. However, it was considered as a method of saving time with more storage spaces. It is suggested in this paper that all frequent item sets of several minimal supports can be stored in a table with a little additional storage, and a representation model is given. Based on this model, the paper systematically discusses the problem of incremental maintenance based on discovered association rules of several minimal supports. Theoretical analysis and experiments show that the approach makes full use of the previous results and reduces the complexity of incremental maintenance algorithms. [ABSTRACT FROM AUTHOR]

In this paper we describe a security system against malicious agents in a wireless ad hoc network. This system monitors the activities of mobile agents. By mining the log data, the system builds dynamic behavioral profiles for each class of agents. The profiles are then used to classify each agent instance and detect anomalous agents. The malicious nature of identified anomalous agents are further determined by a threat assessment module. For mobile agent profiling, we presented two approaches: one based on rule-learning and another on histograms of features related each agent class. We have implemented a prototype system using the histogram-based approach. The prototype is also described in the paper. [ABSTRACT FROM AUTHOR]

Kernel matching pursuit (KMP) is a relatively new learning algorithm to produce non-linear version of conventional supervised and unsupervised learning algorithm. However, it also contains some defects such as storage problem (in training process) and sparsity problem. In this paper, a new method is proposed to pre-select the base vectors from the original data according to vector correlation principle, which could greatly reduce the scale of the optimization problem and improve the sparsity of the solution. The method could capture the structure of the data space by approximating a basis of the subspace of the data; therefore, the statistical information of the training samples is preserved. In the paper, the deduction of mathematical process is given in details and the number of simulation results on artificial data and practical data has been done to validate the performance of base vector selection (BVS) algorithm. The experimental results show the combination of such algorithm with KMP can make great progress while leave the performance almost unchanged. [ABSTRACT FROM AUTHOR]

A new approach is needed to handle huge dataset stored in multiple tables in a very-large database. Data mining and Knowledge Discovery in Databases (KDD) promise to play a crucial role in the way people interact with databases, especially decision support databases where analysis and exploration operations are essential. In this paper, we present related works in Relational Data Mining, define the basic notions of data mining for decision support and the types of data aggregation as a means of categorizing or summarizing data. We then present a novel approach to relational domain learning to support the development of decision making models by introducing automated construction of hierarchical multi-attribute model for decision making. We will describe how relational dataset can naturally be handled to support the construction of hierarchical multi-attribute model by using relational aggregation based on pattern's distance. In this paper, we presents the prototype of "Dynamic Aggregation of Relational Attributes" (hence called DARA) that is capable of supporting the construction of hierarchical multi-attribute model for decision making. We experimentally show these results in a multi-relational domain that shows higher percentage of correctly classified instances and illustrate set of rules extracted from the relational domains to support decision-making. [ABSTRACT FROM AUTHOR]

Advance of high-throughput technologies, such as the microarray and mass spectrometry, has provided an effective approach for the development of systems biology, which aims at understanding the complex functions and properties of biological systems and processes. Revealing the functional correlated genes with co-expression pattern from microarray data allows us to infer the transcriptional regulatory networks and perform functional annotation of genes, and has become one vital step towards the implementation of integrative systems biology. Clustering is particularly useful and preliminary methodology for the discovery of co-expressed genes, for which many conventional clustering algorithms developed in the literature can be potentially useful. However, due to existing large amount of noise and a variety of uncertainties in the microarray data, it is vital important to develop techniques which are robust to noise and effective to incorporate user-specified objectives and preference. For this particular purpose, this paper presented a Genetic Algorithm (GA) based hybrid method for the co-expression gene discovery, which intends to extract the gene groups that have maximal dissimilarity between groups and maximal similarity within a group. The experimental results show that the proposed algorithm is able to extract more meaningful, sensible and significant co-expression gene groups than the traditional clustering methods such as the K-means algorithm. Besides presenting the proposed hybrid GA-based clustering algorithm for co-expression gene discovery, this paper introduces a new framework of integrative systems biology employed in our current research. [ABSTRACT FROM AUTHOR]

In the last decade there has been an explosion of interest in mining time series data, introducing new algorithms to index, classify, cluster and segment time series. In this paper we use fractal theory and reconstructed phase space to analysis the special time series -power systems disturbance signal. After analyzing the feasible method of time series data mining-fractal theory and reconstructed phase space, which is used for the analysis of power disturbance signals. Eight common happed disturbances are considered in this paper, the simulation results show that fractal method can detect the transient disturbance, accurately locate the time when it occurred. Reconstructed phase space can classify the different type of disturbance. It is concluded that two methods are all efficient and intuitionistic for detection and diagnosis the fault of power system, which presents a new concept for power disturbance analysis. [ABSTRACT FROM AUTHOR]

Information Bottleneck method can be used as a dimensionality reduction approach by grouping "similar" features together [1]. In application, a natural question is how many "features groups" will be appropriate. The dependency on prior knowledge restricts the applications of many Information Bottleneck algorithms. In this paper we alleviate this dependency by formulating the parameter determination as a model selection problem, and solve it using the minimum message length principle. An efficient encoding scheme is designed to describe the information bottleneck solutions and the original data, then the minimum message length principle is incorporated to automatically determine the optimal cardinality value. Empirical results in the documentation clustering scenario indicates that the proposed method works well for the determination of the optimal parameter value for information bottleneck method. [ABSTRACT FROM AUTHOR]

As technologies develop in faster and more complicated ways, it is getting more important to expect the direction of technological progresses. So many methods are being proposed all around world and one of them is to use patent information. Moreover, with efforts of governments in many countries, many patent analysis methods have been exploited and suggested usually on the basis of patent documents. However, current patent analysis methods have some limitations. In this paper, we suggest a new visualization method for a patent map, which represents patent analysis results with considering both structured and unstructured items of each patent document. And by the adoption of the k-means clustering algorithm and semantic networks, we suggest concrete steps to make a patent map which gives a clear and instinctive insight on the targeted technology. In application, we built up a patent map for the ubiquitous computing technology and discussed an overall view of its progresses. [ABSTRACT FROM AUTHOR]

Practical pattern classification and knowledge discovery problems require selecting a useful subset of features from a much larger set to represent the patterns to be classified. Exhaustive evaluation of possible feature subsets is usually infeasible in practice because of the large amount of computational effort required. Bio-inspired algorithms offer an attractive approach to find near-optimal solutions to such optimization problems. This paper presents an approach to feature subset selection using bio-inspired algorithms. Our experiments with several benchmark real-world pattern classification problems demonstrate the feasibility of this approach to feature subset selection in the automated design of neural networks for pattern classification and knowledge discovery. [ABSTRACT FROM AUTHOR]

In order to exploit the dependencies in relational data to improve predictions, relational classification models often need to make simultaneous statistical judgments about the class labels for a set of related objects. Robustness has always been an important concern for such collective classification models since many real-world relational data such as Web pages are often accompanied with much noisy information. In this paper, we propose a contextual dependency network (CDN) model for classifying linked objects in the presence of noisy and irrelevant links. The CDN model makes use of a dependency function to characterize the contextual dependencies among linked objects so that it can effectively reduce the effect of irrelevant links on the classification. We show how to use the Gibbs inference framework over the CDN model for collective classification of multiple linked objects. The experiments show that the CDN model demonstrates relatively high robustness on datasets containing irrelevant links. [ABSTRACT FROM AUTHOR]

With the growing deployment of host and network intrusion detection systems, managing the reports from these systems become critically important. Current intrusion detection systems focus on low-level attacks or anomalies. As a result, it is difficult for users or intrusion response systems to understand the intrusion behind the alerts and take appropriate actions. In this paper, we propose alert correlation analysis based on data mining techniques for the management of alerts. Because data mining tasks deal with the discovery of implicit data, we can discover the interconnection and inter relationships among the alerts. So the results of analyzing the alert data are used for the security policy server to construct the security policy rule efficiently in the framework of PBNM(Policy Based Network Management). It helps not only to manage the fault users and hosts but also to discover possible alert sequences. [ABSTRACT FROM AUTHOR]

Comprehensibility is vital in results of medical data mining systems since doctors simply require it. Another important issue specific to some data sets, like Fitness, is their uniform distribution due to tile analysis that was performed on them. In this paper, we propose a novel data mining tool named OSDM (Optimized Shape Distribution Method) to give a comprehensive view of correlations of attributes in cases of uneven frequency distribution among different values of symptoms. We apply OSDM to explore the relationship of the Fitness data and symptoms in medical test dataset for which popular data mining methods fail to give an appropriate output to help doctors decisions. In our experiment, OSDM found several useful relationships. [ABSTRACT FROM AUTHOR]

This paper uses measures of structural cohesion from social network analysis (SNA) literature to discuss how to destabilize terrorist networks by visualizing participation index of various terrorists in the dataset. Structural cohesion is defined as the minimum number of terrorists, who if removed from the group, would disconnect the group. We tested bottom-up measures from SNA (cliques, n-cliques, n-clans and k-plex) using dataset of 9-11 terrorist network, and found that Mohamed Atta, who was known as ring leader of the plot, participated maximum number of groups generated by the structural cohesion measures. We discuss the results of recently introduced algorithms for constructing hierarchy of terrorist networks, so that investigators can view the structure of non-hierarchical organizations, in order to destabilize terrorist networks. Based upon the degree centrality, eigenvector centrality, and dependence centrality measures, a method is proposed to construct the hierarchical structure of complex networks. It is tested on the September 11, 2001 terrorist network constructed by Valdis Krebs. In addition we also briefly discuss various roles in the network i.e., position role index, which discovers various positions in the network, for example, leaders / brokers and followers. [ABSTRACT FROM AUTHOR]

Distributed Denial of Service (DDoS) attack is the greatest security fear for IT managers. With in no time, thousands of vulnerable computers can flood victim website by choking legitimate traffic. Several specific security measurements are deployed to encounter DDoS problem. Instead of specific solution, a comprehensive DDoS cure is needed which can combat against the previously and upcoming DDoS attack vulnerabilities. Development of such solution requires understanding of all those aspects which can help hacker to activate zombies and launch DDoS attack. In this paper, we comprehensively analyzed the DDoS problem and we proposed a simplified taxonomy to categorize the attack scope and available defense solutions. This taxonomy can help the software developers and security practitioners to understand the common vulnerabilities that encourage the attackers to launch DDoS attack. [ABSTRACT FROM AUTHOR]

In this paper, we propose a method to detect hot event automatically. We use all the web pages from Jan 1st 2005 to Dec 31st 2005, and detect new events by using incremental TF-IDF model and incremental cluster algorithm. Based on analysis of the attributes of events, we propose a method to measure the activity of events, then filter and sort the event according to the activity of events; finally a hot event list can be derived. [ABSTRACT FROM AUTHOR]

Sensor networks are expected to be used for spatial cognition in harsh environments. When an event happens, there will be several sensors detect it and send their reports to a sink, but these data are neither integrated nor reliable. Therefore, it is reasonable to make use of data mining on intermediate nodes to acquire deeper knowledge on an event and cut down the total traffic at the same time. Furthermore, the QoS requests should be considered too. In this paper, we propose a centralized algorithm to achieve optimal traffic management on sensor networks with considering QoS and data fusion. Its efficiency is shown by experiments. [ABSTRACT FROM AUTHOR]

Active learning aims at reducing the number of training examples to be labeled by automatically processing the unlabeled examples, then selecting the most informative ones with respect to a given cost function for a human to label. The major problem is to find the best selection strategy function to quickly reach high classification accuracy. Query-by-Committee (QBC) method of active learning is less computation than other active learning approaches, but its classification accuracy can not achieve the same high as passive learning. In this paper, a new selection strategy for the QBC method is presented by combining Vote Entropy with Kullback-Leibler divergence. Experimental results show that the proposed algorithm is better than previous QBC approach in classification accuracy. It can reach the same accuracy as passive learning with few labeled training examples. [ABSTRACT FROM AUTHOR]

Financial distress prediction is an important research topic in both academic and practical world. This paper proposed a financial distress prediction model based on similarity weighted voting case-based reasoning (CBR), which consists of case representation, similar case retrieval and combination of target class. An empirical study was designed and carried out by using Chinese listed companies' three-year data before special treatment (ST) and adopting leave-one-out and grid-search technique to find the model's good parameters. The experiment result of this model was compared with multi discriminant analysis (MDA), Logit, neural networks (NNs) and support vector machine (SVM), and it was concluded that similarity weighted voting CBR model has very good predictive ability for enterprises which will probably run into financial distress in less than two years, and it is more suitable for short-term financial distress prediction. [ABSTRACT FROM AUTHOR]

Logs of a BitTorrent (BT) site on campus are used to characterize the information sharing properties of BT users. The characteristics of (1) user's publishing and downloading behaviors, (2) the life span of the torrent and (3) user's downloaded, uploaded amount are addressed in the paper. Our main contributions are: (1) the distribution of users publishing characteristics has a flat head which is similar to that of user's information fetch behaviors, but cannot be explained by the well-known fetch-at-most-once principle. An indirect selection model is proposed to explain the phenomenon. (2) The lifetime y of a torrent file is tightly correlated with user's interests x (the number of fetches) (3) Distributions of user's downloaded amount and uploaded amount are essentially different. Approximately the former is exponential but the latter is power-law. The Cobb-Douglas like utility (CDLU) function is applied to study the relationship between them and a simple bound is found in user's CDLU. [ABSTRACT FROM AUTHOR]

With the availability of pattern recognition in digital image processing, it becomes essential to automate the data mining and information processing of graph expression patterns. This paper focuses on mining three-dimensional (3D) patterns of the binocular graphs that are collected through two cameras as the left and right cameras. A new algorithm for the stereo matching and 3D pattern recurring based on data mining is presented. The experimental results show that the algorithm can effectively recur the pattern of scene graphs. [ABSTRACT FROM AUTHOR]

The web with its rapid expansion has become an excellent resource for gathering information and people's opinion. A company owner wants to know who is the competitor, and a customer also wants to know which company provides similar product or service to what he/she is in want of. This paper proposes an approach based on mutual information, which focuses on mining competitors of the entity(such as company, product, person ) from the web. The proposed techniques first extract a set of candidates of the input entity, and then rank them according to the comparability, and finally find and organize the reviews related to both original entity and its competitors. A novel system called "CoDis" based upon these techniques is implemented, which is able to automate the tedious process in a domain-independent and web-scale dynamical manner. In the experiment we use 32 different entities distributed in varied domains as inputs and the CoDis discovers 143 competitors. The experimental results show that the proposed techniques are highly effective. [ABSTRACT FROM AUTHOR]

Financial time series show the non-linear and fractal characters in the process of time-space kinetics evolution. Traditional dimension reduction methods for similarity query introduce the smoothness to data series in some degree. In the case of unknowing the fractal dimension of financial non- stationary time series, the process of querying the similarity of curve figure will be affected to a certain degree. In this paper, an evaluation formula of varying-time Hurst index is established and the algorithm of varying-time index is presented, and a new determinant standard of series similarity is also introduced. The similarity of curve basic figure is queried and measured at some resolution ratio level. In the meantime, the fractal dimension in local similarity is matched. The effectiveness of the method is validated by means of the simulation examples. [ABSTRACT FROM AUTHOR]

The pages on the World Wide Web and their hyperlinks induce a huge directed graph - the Web Graph. Many models have been brought up to explain the static and dynamic properties of the graph. Most of them pay much attention to the pages without considering their essential relations. In fact, Web pages are well organized in Web sites as a tree hierarchy. In this paper, we propose a hierarchical model of Web graph which exploits both link structure and hierarchical relations of Web pages. The analysis of the model reveals many properties about the evolution of pages, sites and the relation among them. [ABSTRACT FROM AUTHOR]

Time series forecasting is an important area in data mining research. Feature preprocessing techniques have significant influence on forecasting accuracy, therefore are essential in a forecasting model. Although several feature preprocessing techniques have been applied in time series forecasting, there is so far no systematic research to study and compare their performance. How to select effective techniques of feature preprocessing in a forecasting model remains a problem. In this paper, the authors conduct a comprehensive study of existing feature preprocessing techniques to evaluate their empirical performance in time series forecasting. It is demonstrated in our experiment that, effective feature preprocessing can significantly enhance forecasting accuracy. This research can be a useful guidance for researchers on effectively selecting feature preprocessing techniques and integrating them with time series forecasting models. [ABSTRACT FROM AUTHOR]

Current sequential pattern mining algorithms often produce a large number of patterns. It is difficult for a user to explore in so many patterns and get a global view of the patterns and the underlying data. In this paper, we examine the problem of how to compress a set of sequential patterns using only K SP-Features(Sequential Pattern Features). A novel similarity measure is proposed for clustering SP-Features and an effective SP-Feature combination method is designed. We also present an efficient algorithm, called CSP(Compressing Sequential Patterns) to mine compressed sequential patterns based on the hierarchical clustering framework. A thorough experimental study with both real and synthetic datasets shows that CSP can compress sequential patterns effectively. [ABSTRACT FROM AUTHOR]

This paper investigates a Time-Series Data Mining (TSDM) Technique based fault diagnostic method for short-switch and open-phase faults in a standard 6-switch inverter fed permanent magnet synchronous motor (PMSM) direct torque control (DTC) drive system. For diagnosing the operating condition of an inverter, the reconstructed phase space (RPS) theory is applied to obtain the special feature consisting in the trajectories of phase currents for healthy and faulty operating conditions. The fuzzy C-mean (FCM) algorithm is used to build a fuzzy membership function, an FCM based ANFIS (FCM-ANFIS) is designed to classify different fault patterns. The proposed method has been studied by simulation using MATLAB; which proves that different operating conditions of PMSM DTC drive can be discovered clearly without background knowledge. [ABSTRACT FROM AUTHOR]

Conventional frequent pattern mining algorithms require users to specify some minimum support threshold. If that specified-value is large, users may lose interesting information. In contrast, a small minimum support threshold results in a huge set of frequent patterns that users may not be able to screen for useful knowledge. To solve this problem and make algorithms more user-friendly, an idea of mining the k-most interesting frequent patterns has been proposed. This idea is based upon an algorithm for mining frequent patterns without a minimum support threshold, but with a k number of highest frequency patterns. In this paper, we propose an explorative mining algorithm, called ExMiner, to mine k-most interesting (i.e. top-k) frequent patterns from large scale datasets effectively and efficiently. The ExMiner is then combined with the idea of "build once mine anytime" to mine top-k frequent patterns sequentially. Experiments on both synthetic and real data show that our proposed methods are more efficient compared to the existing ones. [ABSTRACT FROM AUTHOR]

Recent work has shown improvements in text clustering and classification by integrating conceptual features extracted from background knowledge. In this paper we address the problem of text classification with labeled data and unlabeled data. We propose a Latent Bayes Ensemble model based on word-concept mapping and transductive boosting method. With the knowledge extracted from ontologies, we hope to improve the classification accuracy even with large amounts of unlabeled documents. We conducted several experiments on two well-known corpora and the results are compared with Naïve Bayes and TSVM classifiers. [ABSTRACT FROM AUTHOR]

Activity data accumulated in real life, e.g. in terrorist activities and fraudulent customer contacts, presents special structural and semantic complexities. However, it may lead to or be associated with significant business impacts. For instance, a series of terrorist activities may trigger a disaster to the society, large amounts of fraudulent activities in social security program may result in huge government customer debt. Mining such data challenges the existing KDD research in aspects such as unbalanced data distribution and impact-targeted pattern mining. This paper investigates the characteristics and challenges of activity data, and the methodologies and tasks of activity mining. Activity mining aims to discover impact-targeted activity patterns in huge volumes of unbalanced activity transactions. Activity patterns identified can prevent disastrous events or improve business decision making and processes. We illustrate issues and prospects in mining governmental customer contacts. Keywords: Activity data, activity mining, impact-targeted mining, unbalanced data. [ABSTRACT FROM AUTHOR]

Naïve Bayesian Tree is a high-accuracy classification method by combining decision tree and naïve Bayes together. It uses averaged global accuracy as the measurement of goodness in the induction process of the tree structure, and chooses the local classifier that is most specific for the target instance to make the decision. This paper mainly introduces a pruning strategy based on local accuracy estimation. Instead of directly using the most specific local classifier (mostly the classifier in a leaf node) to making classification in NBTree, our pruning strategy uses the measurement of local accuracy to guide the selection of local classifier for decision. Experimental results manifest that this pruning strategy is effective, especially for the NBTree with relatively more nodes. [ABSTRACT FROM AUTHOR]

Community structure is an important topological property of network. Being able to discover it can provide invaluable help in exploiting and understanding complex networks. Although many algorithms have been developed to complete this task, they all have advantages and limitations. So the issue of how to detect communities in networks quickly and correctly remains an open challenge. Distinct from the existing works, this paper studies the community structure from the view of network evolution and presents a self-organizing network evolving algorithm for mining communities hidden in complex networks. Compared with the existing algorithm, our approach has three distinct features. First, it has a good classification capability and especially works well with the networks without well-defined community structures. Second, it requires no prior knowledge and is insensitive to the build-in parameters. Finally, it is suitable for not only positive networks but also singed networks containing both positive and negative weights. [ABSTRACT FROM AUTHOR]

Feature selection, the process of selecting a feature subset from the original feature set, plays an important role in a wide variety of contexts such as data mining, machine learning, and pattern recognition. Recently, fractal dimension has been exploited to reduce the dimensionality of the data space. FDR(Fractal Dimensionality Reduction) is one of the most famous fractal dimension based feature selection algorithm proposed by Traina in 2000. However, it is inefficient in the high dimensional data space for multiple scanning the dataset. Take advantage of the Z-ordering technique, this paper proposed an optimized FDR, ZBFDR(Z-ordering Based FDR), which can select the feature subset through scanning the dataset once except for preprocessing. The experimental results show that ZBFDR algorithm achieves better performance. [ABSTRACT FROM AUTHOR]

Clustering is an important approach to graph partitioning. In this process a graph model expressed as the pairwise similarities between all data objects is represented as a weighted graph adjacency matrix. The min-cut bipartitioning problem is a fundamental graph partitioning problem and is NP-Complete. In this paper, we present an effective multi-level algorithm for bisecting graph. The success of our algorithm relies on exploiting both Tabu search theory and the concept of the graph core. Our experimental evaluations on 18 different graphs show that our algorithm produces excellent solutions compared with those produced by MeTiS that is a state-of-the-art partitioner in the literature. [ABSTRACT FROM AUTHOR]

In the field of data mining (DM), feature selection is one of the basic strategies handling with high-dimensionality problems. This paper makes a review of current methods of feature selection and proposes a unified strategy of feature selection, which divides overall procedures of feature selection into two stages, first to determine the FIF (Feature Important Factor) of features according to DM tasks, second to select features according to FIF. For classifying problems, we propose a new method for determining FIF based on decision trees and provide practical suggestion for feature selection. Through analysis on experiments conducted on UCI datasets, such a unified strategy of feature selection is proven to be effective and efficient. [ABSTRACT FROM AUTHOR]

Complex network is an active research field in complex system in recent years. In this paper, we investigate the topological structure of complex networks and present a novel unsupervised visual clustering algorithm for finding community in complex networks. We firstly introduce a new distance between nodes to measure the dissimilarity between nodes and obtain the distance matrix. Then the rows (columns) of distance matrix are reordered according to the dissimilarity and the reordered matrix is displayed as an intensity image. Clusters are indicated by dark blocks of pixels along the main diagonal. The experiments show that our algorithm has good performance and can find the community structure hidden in complex networks. [ABSTRACT FROM AUTHOR]

One of buzzwords for modern manufacturing industry are flexible manufacturing systems (FMS), in which several machines are interlinked by an automated information and material flow system. Description and control upon these systems are of prominent significance. This paper is concerned with mining and construction of the established FMS from work event logs. A novel Petri nets based algorithm is developed to implement such an idea. When an FMS is mined and constructed, its corresponding Petri net is used to evaluate, analyze, and control the system. Theoretical and experimental results are illustrated to show the effectiveness and efficiency of this approach. [ABSTRACT FROM AUTHOR]

In this paper, we explore a novel type of gene cluster called local conserved gene cluster or LC-Cluster for short. A gene's expression level is local conserved if it is expressed with the similar abundance only on a subset of conditions instead of on all the conditions. A subset of genes which are simultaneously local conserved across the same subset of samples form an LC-Cluster, where the samples correspond to some phenotype and the genes suggest all candidates related to the phenotype. Two efficient algorithms, namely FALCONER and E-FALCONER, are proposed to mine the complete set of maximal LC-Clusters. The test results from both real and synthetic datasets confirm the effectiveness and efficiency of our approaches. [ABSTRACT FROM AUTHOR]

This paper presents a novel genetic algorithm (GA) for multiple sequence alignment in protein analysis. The most significant improvement afforded by this algorithm results from its use of segment profiles to generate the diversified initial population and prevent the destruction of conserved regions by crossover and mutation operations. Segment profiles contain rich local information, thereby speeding up convergence. Secondly, it introduces the use of the norMD function in a genetic algorithm to measure multiple alignment Finally, as an approach to the premature problem, an improved progressive method is used to optimize the highest-scoring individual of each new generation. The new algorithm is compared with the ClustalX and T-Coffee programs on several data cases from the BAliBASE benchmark alignment database. The experimental results show that it can yield better performance on data sets with long sequences, regardless of similarity. [ABSTRACT FROM AUTHOR]

Quantum-behaved Particle Swarm Optimization (QPSO) is a novel optimization algorithm proposed in the previous work. Compared to the original Particle Swarm Optimization (PSO), QPSO is global convergent, while the PSO is not. This paper focuses on exploring the applicability of the QPSO to data clustering. Firstly, we introduce the K-means clustering algorithm and the concepts of PSO and QPSO. Then we present how to use the QPSO to cluster data vectors. After that, experiments are implemented to compare the performance of various clustering algorithms. The results show that the QPSO can generate good results in clustering data vectors with tolerable time consumption. [ABSTRACT FROM AUTHOR]

Radial basis function (RBF) models have been successfully employed to study a broad range of data mining problems and benchmark data sets for real world scientific and engineering applications. In this paper we investigate RBF models with Gaussian kernels by developing classifiers in a systematic way. In particular, we employ our newly developed RBF design algorithm for a detailed performance study and sensitivity analysis of the classification models for the popular Monk's problems. The results show that the accuracy of our classifiers is very impressive while our classification approach is systematic and easy to implement. In addition, differing complexity of the three Monk's problems is clearly reflected in the classification error surfaces for test data. By exploring these surfaces, we acquire better understanding of the data mining classification problems. Finally, we study the error surfaces to investigate trade-offs between different choices of model parameters to develop efficient and parsimonious models for a given application. [ABSTRACT FROM AUTHOR]

In this paper, we propose a multiple-level image semantics classification method. The multiple-level image semantics classifier is constructed according to a hierarchical semantics tree. A semantics tree is defined according to the individual user's habit of managing files. So it is personalized. The classification features are selected by calculating information entropy of images. The hierarchical classifier is constructed according to a class correlation measure. This measure considers both the relation of the classifiers between different hierarchical levels and the relation between the classifiers at the same level. The unlabelled pictures can be classified top-down and assigned to corresponding class and semantic labels. In our experiment binary SVM is used. The hierarchical classifier is built by selecting meta-classifiers with the combinations that have better performance. The result shows that the hierarchical classifier is more effective than a flat method. [ABSTRACT FROM AUTHOR]

This paper introduces a novel classification method-transductive total least square classification method (TTLSC). While inductive approaches are concerned with the development of a model to approximate data in the whole problem space (induction), and consecutively - using this model to calculate the output value(s) for a new input vector (deduction), in transductive systems a local model is developed for every new input vector, based on some closest data to this vector from the training data set. The total least square method (TLS) is one of the optimal fitting methods that can be used for curve and surface fitting and outperform the commonly used least square fitting methods in resisting both normal noise and outlier. The TTLSC is illustrated by a case study: a real medical decision support problem of estimating the survival of haemodialysis patients. This personalized modelling can also be applied to solve other classification or clustering problems. [ABSTRACT FROM AUTHOR]

The issue of maintaining privacy in frequent itemset mining has attracted considerable attentions. In most of those works, only distorted data are available which may bring a lot of issues in the data-mining process. Especially, in the dynamic update distorted database environment, it is nontrivial to mine frequent itemsets incrementally due to the high counting overhead to recompute support counts for itemsets. This paper investigates such a problem and develops an efficient algorithm SA-IFIM for incrementally mining frequent itemsets in update distorted databases. In this algorithm, some additional information is stored during the earlier mining process to support the efficient incremental computation. Especially, with the introduction of supporting aggregate and representing it with bit vector, the transaction database is transformed into machine oriented model to perform fast support computation. The performance studies show the efficiency of our algorithm. [ABSTRACT FROM AUTHOR]

In content-based image retrieval, learning from users' feedback can be considered as an one-class classification problem. However, the OCIB method proposed in [1] suffers from the problem that it is only a one-mode method which cannot deal with multiple interest regions. In addition, it requires a pre-specified radius which is usually unavailable in real world applications. This paper overcomes these two problems by introducing ensemble learning into the OCIB method: by Bagging, we can construct a group of one-class classifiers which emphasize various parts of the data set; this is followed by a rank aggregating with which results from different parameter settings are incorporated into a single final ranking list. The experimental results show that the proposed I-OCIB method outperforms the OCIB for image retrieval applications. [ABSTRACT FROM AUTHOR]

This paper present a hybrid numeric method that integrates a Bayesian statistical method for electricity price spikes classification determination and a Bayesian expert (BE) is described for data mining with experience decision analysis approach. The combination of experience knowledge and support vector machine (SVM) modeling with a Bayesian classification, which can classify the spikes and normal electricity prices, are developed. Bayesian prior distribution and posterior distribution knowledge are used to evaluate the performance of parameters in the SVM models. Electricity prices of one regional electricity market (REM) in China are used to test the proposed method, experimental results are shown. [ABSTRACT FROM AUTHOR]

Mutually associated pattern mining can find such type of patterns whose any two sub-patterns are associated. However, like frequent pattern mining, when the minimum association threshold is set to be low, it still generates a large number of mutually associated patterns. The huge number of patterns produced not only reduces the mining efficiency, but also makes it very difficult for a human user to analyze in order to identify interesting/useful ones. In this paper, a new task of maximal frequent mutually associated pattern mining is proposed, which can dramatically decrease the number of patterns produced without information loss due to the downward closure property of the association measure and meanwhile improve the mining efficiency. Experimental results show that maximal frequent mutually associated pattern mining is quite a necessary approach to lessening the number of results and increasing the performance of the algorithm. Also, experimental results show that the techniques developed are much effective especially for very large and dense datasets. [ABSTRACT FROM AUTHOR]

This paper proposes a polarimetric synthetic aperture radar (PolSAR) data classification method which applies multi-dimensional transform to identify density peaks and valleys for polarimetric signatures clustering. The new approach firstly introduces an improved maximum homogeneous region filter which can effectively preserve structure feature and polarimetric signatures. Then polarimetric signatures are extracted based on Freeman-Durden three-component composition. Finally, we obtain the classification results by multi-dimensional watershed clustering on the extracted polarimetric signatures. The effectiveness of this classification scheme is demonstrated using the full polarimetric L-band SAR imagery. [ABSTRACT FROM AUTHOR]