Intelligent Data Analysis: An International Journal

In this paper we investigate the performance of pairwise (or round robin) classification, originally a technique for turning multi-class problems into two-class problems, as a general ensemble technique. In particular, we show that the use of round robin ensembles will also increase the classification performance of decision tree learners, even though they can directly handle multi-class problems. The performance gain is not as large as for bagging and boosting, but on the other hand round robin ensembles have a clearly defined semantics. Furthermore, we investigate whether confidence estimates can be used to improve the accuracy of the predictions of the ensemble. Finally, we show that the advantage of pairwise classification over direct multi-class classification and one-against-all binarization increases with the number of classes, and that round robin ensembles form an interesting alternative for problems with ordered class values.

This paper introduces non-Euclidean c-means clustering algorithms. These algorithms rely on weighted norms to measure the distance between the feature vectors and the prototypes that represent the clusters. The proposed algorithms are developed by solving a constrained minimization problem in an iterative fashion. The norm weights are determined from the data in an attempt to produce partitions of the feature vectors that are consistent with the structure of the feature space. A series of experiments on three different data sets reveal that the proposed non-Euclidean c-means algorithms provide an attractive alternative to Euclidean c-means clustering in applications that involve data sets containing clusters of different shapes and sizes.

The k-Nearest Neighbour (k-NN) method is a typical lazy learning paradigm for solving classification problems. Although this method was originally proposed as a non-parameterised method, attribute weight setting has been commonly adopted to deal with irrelevant attributes. In this paper, we propose a new attribute weight setting method for k-NN based classifiers using quadratic programming, which is particularly suitable for binary classification problems. Our method formalises the attribute weight setting problem as a quadratic programming problem and exploits commercial software to calculate attribute weights. To evaluate our method, we carried out a series of experiments on six established data sets. Experiments show that our method is quite practical for various problems and can achieve a stable increase in accuracy over the standard k-NN method as well as a competitive performance. Another merit of the method is that it can use small training sets.

Weighted Local Similarity Pattern (WLSP)} is proposed as a new image similarity model, which considers two fundamental properties of the human visual system: (1) saliency of regions within an image, and (2) saliency of features within each region. Furthermore, since both region and feature saliencies are context dependent, genetic algorithm (GA)-based relevance feedback mechanism is proposed to automatically infer the (sub-)optimal assignment of the two saliencies, based on the query image and the set of relevant images, provided by the user. None of the existing image similarity models considers both region and feature saliencies in a context-dependent sense, allowing their automatic inference. In addition, this paper is the first to explicitly discuss the implications of the region and feature saliency properties to the design of an image similarity model, in the framework of image retrieval. The proposed method – including the WLSP image similarity model and the GA-based relevance feedback mechanism – is evaluated on five test databases, with around 2,500 images, covering 62 semantic categories. Compared with eleven of the representative image similarity models, including three based on relevance feedback, the proposed model brings in average between 6% and 30% increase in the retrieval precision. Results suggest that considering region and feature saliencies in a context-dependent sense enables the image similarity model to more accurately capture the human similarity perception.

Typing rhythms are the rawest form of data stemming from the interaction between users and computers. When properly sampled and analyzed, they may become a useful tool to ascertain personal identity. Moreover, unlike other biometric features, typing dynamics have an important characteristic: they still exist and are available even after an access control phase has been passed. As a consequence, keystroke analysis can be used as a viable tool for user authentication throughout the work session.

In this paper we present an original approach to identity verification based on the analysis of the typing rhythms of individuals on different texts. Our experiments involve 130 volunteers and reach the best outcomes found in the literature, using a smaller amount of information than in other works, and avoiding any form of tailoring of the system to the available data set. The method described in the paper is easily tuned to reach an acceptable trade-off between the need to spot most impostors and to avoid false alarms, and, as a consequence, it can become a valid aid to intrusion detection.