A Pre-processing Study to Solve the Problem of Rare Class Classification of Network Traffic Data

In order to ensure the healthy and orderly development of enterprises, it is of great importance to effectively predict enterprise financial risk in advance, and then provide early warning information to enterprise’s managers.Firstly, we design an index system for enterprise financial risk early warning, which contains five aspects: 1) Profitability, 2) Debt paying ability, 3) Operation ability, 4) Growth ability, and 5) Non-financial indicators. Secondly, we propose a novel hybrid PSO-SVM model based enterprise financial risk early warning algorithm by converting the proposed problem to a classification problem. As the performance of SVM classifier highly relies on parameter selection, we introduce the PSO algorithm to estimate optimal parameters for SVM. Thirdly, we choose several ST companies of the listed companies in China as financial crisis enterprises, and compare their running states with Non-ST companies. Main contributions of this paper lie in that we propose a hybrid PSO-SVM model based enterprise financial risk early warning algorithm by solving a classification problem. Experimental results show that the proposed algorithm is able to effectively provide early warning information for enterprise financial risk, and performs better than BP neural network and SVM without parameter optimization.

Crow Search Algorithm is a novel Metaheuristic method based on the intelligent behavior of crows. It has been used to solve some optimization problems like engineering design problems, feature extraction and classification problems but has not been applied to vehicle routing problem. In this paper, crow search algorithm has been utilized to solve capacitated vehicle routing problem and the performance of it on various size-capacitated vehicle routing problem instances is analyzed. The factorial design ANOVA is used to determine the performance of crow search algorithm under different parameter settings on capacitated vehicle routing problem instances.

The classification problems described in the Machine Learning literature usually relate to the classification of data in which each example is associated to a class belonging to a finite set of classes, all at the same level. However, there are classification issues, of a hierarchical nature, where the classes can be either subclasses or super classes of other classes. In many hierarchical problems, one or more examples may be associated with more than one class simultaneously. These problems are known as hierarchical multi-label classification (HMC) problems. In this work, the ML-KNN algorithm was used to predict hierarchical multi-label problems, in order to determine the number of classes that can be assigned to an example. Through the experiments performed on 10 protein function databases and the statistical analysis of the results, it can be shown that the adaptations performed in the ML-KNN algorithm brought significant performance improvements based on the hierarchical precision and recall metrics Hierarchical.

In this paper the use of neural networks in the solution of a classification problem is described and a non parametric classifier which combines radial basis function networks (RBF networks) and binary classification trees is proposed. It is shown that from a classification tree a class of RBF networks can be derived, whose elements differ in the degree of dependence from the discriminant rules described by the tree. A simulation study based on a two-class problem shows that the proposal represents a valid solution, particularly in presence of noisy variables.

Horizontal Gene Transfer (HGT), a process of acquisition and fixation of foreign genetic material, is an important biological phenomenon. Several approaches to HGT inference have been proposed. However, most of them either rely on approximate, non-phylogenetic methods or on the tree reconciliation, which is computationally intensive and sensitive to parameter values. In this work, we investigate the Locus Tree Inference problem as a possible alternative that combines the advantages of both approaches. We show several algorithms to solve the problem in the parsimony framework. We introduce a novel tree mapping, which allows us to obtain a heuristic solution to the problems of locus tree inference and duplication classification. Our approach allows not only for faster comparisons of gene and species trees but also to improve known algorithms for duplication inference in the presence of polytomies in the species trees.

User privacy is widely affected by the occurrence of privacy-invasive software (PIS) on the Internet. Various forms of countermeasures try to mitigate the negative effects caused by PIS. We use a computer forensic tool to evaluate an anti-spyware tool, with respect to found PIS over a four years period. Within the anti-spyware tool PIS was slowly identified, caused classification problems, and formely classified PIS were sometimes excluded. Background information on both PIS and countermeasure techniques are also presented, followed by discussions on legal disputes between developers of PIS and vendors of countermeasures.

As the amount of data acquired from humans is constantly increasing, efficient tools are needed for extracting relevant information from this data. This paper presents a Matlab implementation of a method to classify and visually explore (highly) multi-variate patient data. The method uses the so-called Disease State Index (DSI) which measures the fit of a test subject's data to two classes present in the data (e.g. 'controls' and 'positives'). DSI values of the different variables measured from a patient can be combined and visualized in a tree-like form using the Disease State Fingerprint (DSF) method. This allows a researcher to explore and understand the relevance of the different variables in classification problems. Moreover, the method is robust with respect to missing data. After giving an introduction to the DSF and DSI methods, the paper describes the steps required to use the methods and presents a MATLAB toolbox to perform these steps. To demonstrate the methods' versatility, the paper illustrates the usage of the toolbox in a few different contexts in which personal health data is to be classified. With this implementation, a powerful and flexible tool is made available to the biomedical research community.

In dynamically control field, the effectiveness of closed loop control is based on good knowledge of system parameters; it is then important to detect parameters changes, chiefly in off-line estimation field. According to the physical structure of plants represented by closed loop autoregressive models the system parameters may be divided into four classes: level and dynamic of transform system, level and dynamic of control system. This gives rise to four kinds of parameter ruptures, where rupture means step-change from one state to another. In the present paper the on-line rupture detection and classification problem is solved by means of a multiple decision procedure which enables the user to localize the rupture area.

Решение обеих проблем хорошо известно в случае общего положения, когда среди собственных чисел линейной части нет чисто мнимых. В этом случае топологический тип особой точки определяется количествами собственных чисел в левой и в правой полуплоскости. Поэтому существуют алгебраические критерии, позволяющие определить топологический тип особой точки общего положения конечным числом арифметических действий над коэффициентами Тейлора правой части дифференциального уравнения.

Publication in the conference proceedings of EUSIPCO, Poznan, Poland, 2007

A novel approach for instance selection in classification problems is presented. This adaptive instance selection is designed to simultaneously decrease the amount of computation resources required and increase the classification quality achieved. The approach generates new training samples during the evolutionary process and changes the training set for the algorithm. The instance selection is guided by means of changing probabilities, so that the algorithm concentrates on problematic examples which are difficult to classify. The hybrid fuzzy classification algorithm with a self-configuration procedure is used as a problem solver. The classification quality is tested upon 9 problem data sets from the KEEL repository. A special balancing strategy is used in the instance selection approach to improve the classification quality on imbalanced datasets. The results prove the usefulness of the proposed approach as compared with other classification methods.

This thesis proposes a novel unified boosting framework. We apply this framework to the several face processing tasks, face detection, facial feature localisation, and pose classification, and use the same boosting algorithm and the same pool of features (local binary features). This is in contrast with the standard approaches that make use of a variety of features and models, for example AdaBoost, cascades of boosted classifiers and Active Appearance Models. The unified boosting framework covers multivariate classification and regression problems and it is achieved by interpreting boosting as optimization in the functional space of the weak learners. Thus a wide range of smooth loss functions can be optimized with the same algorithm. There are two general optimization strategies we propose that extend recent works on TaylorBoost and Variational AdaBoost. The first proposition is an empirical expectation formulation that minimizes the average loss and the second is a variational formulation that includes an additional penalty for large variations between predictions. These two boosting formulations are used to train real-time models using local binary features. This is achieved using look-up-tables as weak learners and multi-block Local Binary Patterns as features. The resulting boosting algorithms are simple, efficient and easily scalable with the available resources. Furthermore, we introduce a novel coarse-to-fine feature selection method to handle high resolution models and a bootstrapping algorithm to sample representative training data from very large pools of data. The proposed approach is evaluated for several face processing tasks. These tasks include frontal face detection (binary classification), facial feature localization (multivariate regression) and pose estimation (multivariate classification). Several studies are performed to assess different optimization algorithms, bootstrapping parametrizations and feature sharing methods (for the multivariate case). The results show good performance for all of these tasks. In addition to this, two other contributions are presented. First, we propose a context-based model for removing the false alarms generated by a given generic face detector. Second, we propose a new face detector that predicts the Jaccard distance between the current location and the ground truth. This allows us to formulate the face detection problem as a regression task.

Psychophysics with high dimensional stimuli (e.g. video fragments) can be challenging. Typically, a stimulus is presented with a noise mask but for high res stimulus displays, the number of free parameters increases exponentially inducing the curse of dimensionality. Another question is how such noise embeddings and stimulus manipulations affect the observers’ strategy [Macke &Wichmann 2010]. This research aims to investigate the errors we make in every day life in detecting emotional expressions. Opposed to noise based experiments, in every day life, faces are often fully visible and errors emerge rather from the misinterpretation of ambiguous but visible components of the face. Macke and Wichmann (2010) investigated a similar problem in gender classification. They presented static face images in a parameterized morph-space [Blanz & Vetter 1999]. We applied the same idea to classification of dynamic emotional expressions but we used Active Appearance Models [Cootes et al. 1998] to parameterize shape variations by low dimensional vectors of facial components. Via staircase procedures, we varied the amount of information in the different components. Participants detected dynamic emotional expressions ranging from no ambiguity at all, towards a static image with neutral expression. We determined separate thresholds for different facial components and found that participants were equally accurate in detecting emotional expressions as long as the eyes were informative.

The problem of the classification, description by the difference equations and possible models of quasihomogeneous autoregressive random fields, existing in one-dimensional space Rl, is considered. The properties of the quasihomogeneous areas as well as of the parameters changing by not the jumps areas of such fields are considered also. The quasihomogeneous areas determination algorithm is proposed.

Computer vision and pattern recognition systems play an important role in our lives by means of automated face detection, face and gesture recognition, and estimation of gender and age. We have developed a gender classifier with performance superior to existing gender classifiers. This paper addresses the problem of gender classification using frontal facial images. The testbed consists of 500 images (250 females and 250 males) randomly withdrawn from the FERET facial database. Independent component analysis (ICA) is used to represent each image as a feature vector in a low dimensional subspace. Different classifiers are studied in this lower dimensional subspace. Our experimental results show the best accuracy of 96% in gender classification by combining ICA and support vector machines (SVMs).