Search

Clustering is an effective tool that divides data into different classes to reveal internal and previously unknown data schemes. However, in conventional clustering algorithms such as the k-means, k-NN, fuzzy c tool, the selection of the appropriate number of clusters for each data set is uncertain and varies with the data sets. Furthermore, the data sets to which the clustering algorithm is applied generally have nonlinear boundaries between clusters. Determining these nonlinear boundaries in the input space causes a complex problem. To overcome these problems, kernel-based clustering methods have been developed in recent years, which automatically determine the number and boundaries of clusters. In particular, the Support Vector Clustering (SVC) algorithm has received great attention in data analysis because of its features such as automatically determining the number of clusters and recognizing nonlinear boundaries based on the Gaussian kernel parameter. The number of clusters and region boundaries produced by SVC may show variation depending on the choice of the kernel function and its parameters. Therefore, the choice of kernel function plays a significant role. In this study, for the first time, the implementation of two different kernel (Cauchy and Laplacian) functions and evaluation of their performances have been realized within the framework of SVC. It was observed that the Laplacian kernel function performed better than Gauss and Cauchy kernel functions.

We extend a procedure based on support vector clustering and devoted to inferring the membership function of a fuzzy set to the case of a universe of discourse over which several fuzzy sets are defined. The extended approach learns simultaneously these sets without requiring as previous knowledge either their number or labels approximating membership values. This data-driven approach is completed via expert knowledge incorporation in the form of predefined shapes for the membership functions. The procedure is successfully tested on a benchmark.

All drivers have their own habitual choice of driving behavior, causing variations in fuel consumption. It would be beneficial to classify these driving styles and extract the most economical and ecological driving patterns. However, driving style of each driver is not consistent and may vary within a single trip. Therefore, this paper proposes a novel technique to robustly classify driving style using the Support Vector Clustering approach, which attempts to differentiate the variations in individual's driving pattern and provides an objective driver classification. It is part of a research program aiming to replicate some humans' driving behaviors on chassis dynamometer using a robot driver. Moreover, it can potentially be used in developing more economical and personalized advanced driver assistance systems (ADAS) and humanized autonomous driving strategies. With the easily accessible on-board diagnostics (OBD) data on modern vehicles, both vehicle state and traffic information of three drivers were collected using an instrumented vehicle, which had external forward-looking radar and a monocular dashcam. For data processing, each trip data was first segmented into separate event groups. Prominent factors were then extracted by applying Principal Component Analysis (PCA) on both statistical and spectral features of all signals. Afterwards, Support Vector Clustering (SVC) was performed to classify driving style during the trip. The trained classifier was used to indicate the driving pattern variations in percentage. The validity of the proposed method was evaluated using the jerk profile, where a high correlation was found between the classification results and jerk distributions. Moreover, a positive relation between fuel consumption and driving aggressivity was also confirmed. Furthermore, it was found that weather condition, time of the day and ultimately, the driver's eagerness, can cause significant variations in driving style.

We propose a procedure devoted to the induction of a shadowed set through the post-processing of a fuzzy set, which in turn is learned from labeled data. More precisely, the fuzzy set is inferred using a modified support vector clustering algorithm, enriched in order to optimize the fuzziness grade. Finally, the fuzzy set is transformed into a shadowed set through application of an optimal alpha-cut. The procedure is tested on synthetic and real-world datasets.

Support vector clustering (SVC) has been successfully used to achieve robust clustering. SVC has the capability of dealing with noise and outliers, and can identify clusters of arbitrary shapes. Moreover, SVC has its superiorities for automatically obtaining the number of clusters and having no requirement for prior knowledge to determine the system topological structure. However, in most cases, it is not possible to choose the best of input parameter values for an input dataset which restricts the application of SVC. In this paper, we propose a support vector clustering approach termed locally weighted support vector clustering, which tackles the automatic parameter estimation problem for SVC. An exponential weighting to each data point is assigned to reflect the within-class importance of different data points in order to detect outliers and noise efficiently. The experiments on multiple real-world datasets demonstrate the effectiveness of our approach.

Hyper-plane based Support Vector Machine (SVM) for classification and hyper-spherical based Support Vector Data Description (SVDD) for clustering have been shown to be very effective in many real-world machine learning problems in various domains. Support vector clustering (SVC) based on SVDD has been successfully used to achieve robust clustering. However, in most cases, it is not possible to choose the best of input parameter values for an input dataset which restricts the application of SVC. Only a few heuristic approaches have been proposed to set the different parameters of these methods in an unsupervised manner. In this paper, we propose an ensemble based clustering approach to tackle the automatic parameter estimation problem for SVC in a fully unsupervised manner which is capable of producing robust clustering results. The experiments on multiple real-world datasets demonstrate the effectiveness of our approach.

This paper describes a psychophysiological-signal-based clustering framework for detecting the changes in operator mental workload incurred by a simulated process control task. A combination of locally linear embedding and support vector clustering approaches is adopted. The unsupervised method is shown to be able to extract features from several channels of the electroencephalogram (EEG) data and to determine whether or not the level of mental workload changes. Simulation results have also demonstrated that a few data clusters can be derived to interpret the change in the operator workload.

Support vector clustering (SVC) is a versatile clustering technique that is able to identify clusters of arbitrary shapes by exploiting the kernel trick. However, one hurdle that restricts the application of SVC lies in its sensitivity to the kernel parameter and the trade-off parameter. Although many extensions of SVC have been developed, to the best of our knowledge, there is still no algorithm that is able to effectively estimate the two crucial parameters in SVC without supervision. In this paper, we propose a novel support vector clustering approach termed ensemble-driven support vector clustering (EDSVC), which for the first time tackles the automatic parameter estimation problem for SVC based on ensemble learning, and is capable of producing robust clustering results in a purely unsupervised manner. Experimental results on multiple real-world datasets demonstrate the effectiveness of our approach., To appear in ICPR 2016

In this paper, we propose a novel text categorization method based on modified Support Vector Clustering (SVC). SVC is a density based clustering approach, which handles the arbitrary shape clusters effectively. The main drawback of traditional SVC is that it treats unclassified documents as outliers. To overcome this problem, we employed Fuzzy C-Means (FCM) to cluster unclassified documents. The modified SVC (SVC-FCM) is applied to categorize text documents. The proposed method consists of three steps: In the first step, Regularized Locality Preserving Indexing (RLPI) is applied on Term Document Matrix (TDM) to reduce dimensionality of features. In second step, we use SVC to find base-cluster centers of documents. Finally, we use FCM to cluster unclassified documents. To evaluate the performance of the proposed method, we conducted experiments on standard 20-NewsGroup dataset.

The support vector clustering (SVC) algorithm consists of two main phases: SVC training and cluster assignment. The former requires calculating Lagrange multipliers and the latter requires calculating adjacency matrix, which may cause a high computational burden for cluster analysis. To overcome these difficulties, in this paper, we present an improved SVC algorithm. In SVC training phase, an entropy-based algorithm for the problem of calculating Lagrange multipliers is proposed by means of Lagrangian duality and the Jaynes' maximum entropy principle, which evidently reduces the time of calculating Lagrange multipliers. In cluster assignment phase, the kernel matrix is used to preliminarily classify the data points before calculating adjacency matrix, which effectively reduces the computing scale of adjacency matrix. As a result, a lot of computational savings can be achieved in the improved algorithm by exploiting the special structure in SVC problem. Validity and performance of the proposed algorithm are demonstrated by numerical experiments.

A two-phase online anomaly detection method based on support vector clustering (SVC) in the presence of non-stationary data is developed in this paper which permits arbitrary-shaped data clusters to be precisely treated. In the first step, offline learning is performed to achieve an appropriate detection model. Then the current model dynamically evolves to match the rapidly changing real-world data. To reduce the dimension of the quadratic programming (QP) problem emerging in the SVC, self-organizing map (SOM) and a replacement mechanism are used to summarize the incoming data. Thus, the proposed method can be efficiently and effectively useable in real time applications. The performance of the proposed method is evaluated by a simulated dataset, three subsets extracted from the KDD Cup 99 dataset, and the keystroke dynamics dataset. Results illustrate capabilities of the proposed method in detection of new attacks as well as normal pattern changes over the time.

The support vector clustering (SVC) algorithm is a recently emerged unsupervised learning method inspired by support vector machines. One key step involved in the SVC algorithm is the cluster assignment of each data point. A new cluster labeling method for SVC is developed based on some invariant topological properties of a trained kernel radius function. Benchmark results show that the proposed method outperforms previously reported labeling techniques.

Traditional methods for face recognition do not scale well with the number of training sample, which limits the wide applications of related techniques. We propose an improved Support Vector Clustering algorithm to handle the large-scale biometric feature data effectively. We prove theoretically that the proposed algorithm converges to the optimum within any given precision quickly. Compared to related state-of-the-art Support Vector Clustering algorithms, it has the competitive performances on both training time and accuracy. Besides, we use the proposed algorithm to handle classification problem, and face recognition, as well. Experiments on synthetic and real-world data sets demonstrate the validity of the proposed algorithm.

The original publication is available at www.springerlink.com; International audience; This paper addresses the fingerprinting of devices that speak a common, yet unknown to the fingerprinting engine, protocol. We consider a behavioral approach, where the fingerprinting of an unknown protocol is based on detecting and exploiting differences in the observed behavior from two or more devices. Our approach assumes zero knowledge about the syntax and state machine underlying the protocol. The main contribution of this paper consists in a two phased method. The first phase identifies the different message types using an unsupervised support vector clustering algorithm. The second phase is leveraging recent advances in tree support kernel in order to learn and differentiate different implementations of that protocol. The key idea is to represent behavior in terms of trees and learn the distinctive subtrees that are specific to one particular device. Our solution is passive and does not assume active and stimulus triggered behavior templates. We instantiate our solution to the particular case of a VoIP specific protocol (SIP) and validate it using extensive data sets collected on a large size VoIP testbed.

Classification of electroencephalogram (EEG) is an important and challenging issue for brain computer interface (BCI) system. In this paper, an algorithm based on common spatial subspace decomposition (CSSD) and support vector clustering (SVC) is proposed to classify single-trial EEG recording during left or right finger movement. The algorithm is tested by the dataset IV of "BCI competition 2003", and the experimental result shows the proposed method, only using bereitschaftspotential (BP), rather than both BP and event-related desynchronization (ERD), has higher classification accuracy than the best one reported in the competition.

Clustering is the preferred choice of method in many applications, and support vector clustering (SVC) has proven efficient for clustering noisy and high-dimensional data sets. A method for multiscale support vector clustering is demonstrated, using the recently emerged method for fast calculation of the entire regularization path of the support vector domain description. The method is illustrated on artificially generated examples, and applied for detecting blood vessels from high resolution time series of magnetic resonance imaging data. The obtained results are robust while the need for parameter estimation is reduced, compared to support vector clustering.

Modern electronic warfare faces complex and dense pulses environments, which brings a severe challenge to radar signal sorting. Support Vector Clustering (SVC) is used into the radar signal sorting and the notion of type-entropy is presented in this paper. And combine the recognition technology based on type-entropy with the sorting technology based on SVC to constitute a novel radar sequence signal sorting system. The experiment result shows that the system can sort efficiently radar signals in the complex pulses environment.

In this paper, we propose a new approach to detect activated time series in functional MRI using support vector clustering (SVC). We extract Fourier coefficients as the features of fMRI time series and cluster these features by SVC. In SVC, these features are mapped from their original feature space to a very high dimensional kernel space. By finding a compact sphere that encloses the mapped features in the kernel space, one achieves a set of cluster boundaries in the feature space. The SVC is an effective and robust fMRI activation detection method because of its advantages in (1) better discovery of real data structure since there is no cluster shape restriction, (2) high quality detection results without explicitly specifying the number of clusters, (3) the stronger robustness due to the mechanism in outlier elimination. Experimental results on simulated and real fMRI data demonstrate the effectiveness of SVC.

We propose a new approach by combining support vector clustering (SVC) with a two layered neural network for multiclass classification problems. This new approach is not only capable of improving the performance of the traditional SVC, but also, compared to support vector machines (SVM), can reduce the computational complexity and get better performance in the case of multiclass problems. Finally, the experimental results demonstrate the effectiveness and efficiency of our approach.

As an important boundary-based clustering algorithm, support vector clustering (SVC) can benefit many real applications owing to its capability of handling arbitrary cluster shapes, especially those directly or indirectly related to pattern exploration and description. As the application deepens, the importance of performance (i.e. criterions of accuracy and efficiency) of SVC increases. To identify gaps in the current methods and propose novel research directions for SVC, we present a survey of the literature in this area. Our approach is to classify the most recent advances into either theory or application. For theoretical contributions, advances related to parameter selection and optimization, dual-problem solutions, and cluster labeling are introduced. We also simultaneously summarize the advantages and drawbacks of each study. With respect to applications, we clearly describe eight groups of schemes based on SVC, either as individual or hybrid methods. Finally, we identify the gaps in SVC research and suggest several future research issues and trends.

Clustering algorithms have evolved to handle more and more complex structures. However, the measures that allow to qualify the quality of such clustering partitions are rare and have been developed only for specific algorithms. In this work, we propose a new cluster validity measure (CVM) to quantify the clustering performance of hierarchical algorithms that handle overlapping clusters of any shape and in the presence of outliers. This work also introduces a cluster merging system (CMS) to group clusters that share outliers. When located in regions of cluster overlap, these outliers may be issued by a mixture of nearby cores. The proposed CVM and CMS are applied to hierarchical extensions of the Support Vector and Gaussian Process Clustering algorithms both in synthetic and real experiments. These results show that the proposed metrics help to select the appropriate level of hierarchy and the appropriate hyperparameters. HighlightsCluster validity measure for arbitrary shaped clusters with outliers.Cluster merging system grouping cluster cores based on the outliers� structure.Truly hierarchical variants of support vector and Gaussian process clustering.Benefits for unsupervised change detection applications are presented.