semi-supervised clustering

Over recent decades, there has been a growing interest in semi-supervised clustering. Compared to the supervised or unsupervised clustering methods for solving different real-life problems, reviewed articles show that semi-supervised clustering methods are more powerful, and even a small amount of supervised information can significantly improve the results of unsupervised methods. One popular method of incorporating partial supervised information is through labeled data. In this study, we propose a semi-supervised clustering algorithm called ConvexClust. The proposed method improves data clustering using a geometric view borrowed from the Lune concept in the connectivity index and 10% of labeled data. Clustering starts with the use of labeled data and the formation of a convex hull. It continues over the labeling of non-labeled data and the updating of the convex hull in an iterative process. Evaluations of three UCI datasets and sixteen artificial datasets show that the proposed method outperforms the other semi-supervised and traditional clustering techniques.

Machine-learning solutions to classification, clustering and matching problems critically depend on the adopted metric, which in the past was selected heuristically. In the last decade, it has been demonstrated that an appropriate metric can be learnt from data, resulting in superior performance as compared with traditional metrics. This has recently stimulated a considerable interest in the topic of metric learning, especially using kernel functions, which map data to feature spaces with enhanced class separability, and implicitly define a new metric in the original feature space. The formulation of the problem of metric learning depends on the supervisory information available for the task. In this paper, we focus on semi-supervised kernel based distance metric learning where the training data set is unlabelled, with the exception of a small subset of pairs of points labelled as belonging to the same class (cluster) or different classes (clusters). The proposed method involves creating a pool of kernel functions. The corresponding kernels matrices are first clustered to remove redundancy in representation. A composite kernel constructed from the kernel clustering result is then expanded into an orthogonal set of basis functions. The mixing parameters of this expansion are then optimised using point similarity and dissimilarity information conveyed by the labels. The proposed method is evaluated on synthetic and real data sets. The results show the merit of using similarity and dissimilarity information jointly as compared to using just the similarity information, and the superiority of the proposed method over all the recently introduced metric learning approaches.

Network traffic identification is an essential component for effective network analysis and management. Signature-based and machine learning techniques are the two most important methods in network traffic analysis. Due to the strengths and weaknesses of these two approaches, their combination can strengthen them and remove the weaknesses of each in detection process. In this article, a hybrid method is introduced, to identify major network tunneling protocols. This method can detect the well-known tunneling protocols by combining signature-based methods and statistical analysis techniques through a clustering algorithm. In this proposed method, the clustering process is refined by the feedback of signature-base method. Since, in semi-supervised clustering, it is important to gain most informative data to improve the clustering performance, in the proposed clustering method, a new active learning approach is introduced for selecting informative constraints. In this hybrid method, four tunneling protocols (L2TP, PPTP, IPsec and OpenVPN) are applied. The obtained results indicate that this proposed hybrid method significantly increases accuracy and cluster purity, and these protocols are identified with high accuracy and low processing cost.