Parameter optimization of kernel-based one-class classifier on imbalance learning

Compared with conventional two-class learning schemes, one-class classification simply uses a single class in the classifier training phase. Applying one-class classification to learn from unbalanced data set is regarded as the recognition based learning and has shown to have the potential of achieving better performance. Similar to twoclass learning, parameter selection is a significant issue, especially when the classifier is sensitive to the parameters. For one-class learning scheme with the kernel function, such as one-class Support Vector Machine and Support Vector Data Description, besides the parameters involved in the kernel, there is another one-class specific parameter: the rejection rate v. In this paper, we proposed a general framework to involve the majority class in solving the parameter selection problem. In this framework, we first use the minority target class for training in the one-class classification stage; then we use both minority and majority class for estimating the generalization performance of the constructed classifier. This generalization performance is set as the optimization criteria. We employed the Grid search and Experiment Design search to attain various parameter settings. Experiments on UCI and Reuters text data show that the parameter optimized one-class classifiers outperform all the standard one-class learning schemes we examined.

Parameter estimation of one-class SVM on imbalance text classification

Compared with conventional two-class learning schemes, one-class classification simply uses a single class for training purposes. Applying one-class classification to the minorities in an imbalanced data has been shown to achieve better performance than the two-class one. In this paper, in order to make the best use of all the available information during the learning procedure, we propose a general framework which first uses the minority class for training in the one-class classification stage; and then uses both minority and majority class for estimating the generalization performance of the constructed classifier. Based upon this generalization performance measurement, parameter search algorithm selects the best parameter settings for this classifier. Experiments on UCI and Reuters text data show that one-class SVM embedded in this framework achieves much better performance than the standard one-class SVM alone and other learning schemes, such as one-class Naive Bayes, one-class nearest neighbour and neural network.

Ensemble parameter estimation for graphical models

Parameter Estimation is one of the key issues involved in the discovery of graphical models from data. Current state of the art methods have demonstrated their abilities in different kind of graphical models. In this paper, we introduce ensemble learning into the process of parameter estimation, and examine ensemble parameter estimation methods for different kind of graphical models under complete data set and incomplete data set. We provide experimental results which show that ensemble method can achieve an improved result over the base parameter estimation method in terms of accuracy. In addition, the method is amenable to parallel or distributed processing, which is an important characteristic for data mining in large data sets.

Can we obtain realistic parameter estimates for the 'protection for sale' model?

In the Grossman and Helpman (1994) model of endogenous trade protection, sectoral lobbies try to influence an incumbent government that maximizes a weighted sum of political contributions and aggregate welfare. We empirically investigate this model using U.S. and Turkish data. Our specification is more tightly tied to theory than those in existing studies. Additionally, we assume all specific‐factor owners to be organized into different lobbies. These changes, validated by hypothesis tests, yield more realistic parameter estimates of the government's concern for aggregate welfare and of the fraction of population organized into lobbies.

A new modelling approach to simulate preferential flow and transport in water repellent media: parameter sensitivity, and effects on crop growth and solute leaching

A modified version of the popular agrohydrological model SWAP has been used to evaluate modelling of soil water flow and crop growth at field situations in which water repellency causes preferential flow. The parameter sensitivity in such situations has been studied. Three options to model soil water flow within SWAP are described and compared: uniform flow, the classical mobile-immobile concept, and a recent concept accounting for the dynamics of finger development resulting from unstable infiltration. Data collected from a severely water-repellent affected soil located in Australia were used to compare and evaluate the usefulness of the modelling options for the agricultural management of such soils.

The study shows that an assumption of uniform flow in a water-repellent soil profile leads to an underestimation of groundwater recharge and an overestimation of plant transpiration and crop production. The new concept of modelling taking finger dynamics into account provides greater flexibility and can more accurately model the observed effects of preferential flow compared with the classical mobile–immobile concept. The parameter analysis indicates that the most important factor defining the presence and extremity of preferential flow is the critical soil water content.

Comparison of the modelling results with the Australian field data showed that without the use of a preferential flow module, the effects of the clay amendments to the soil were insufficiently reproduced in the dry matter production results. This means that the physical characteristics of the soil alone are not sufficient to explain the measured increase in production on clay amended soils. However, modelling with the module accounting for finger dynamics indicated that the preferential flow in water repellent soils that had not been treated with clay caused water stress for the crops, which would explain the decrease in production.

A parameter adjustment method for relevance feedback

Various relevance feedback techniques have been applied in Content-Based Image Retrieval (CBIR). By using relevance feedback, CBIR allows the user to progressively refine the system's response to a query. In this paper, after analyzing the feature distributions of positive and negative feedbacks, a new parameter adjustment method for iteratively improving the query vector and adjusting the weights is proposed. Experimental results demonstrate the effectiveness of this method.

Intelligent assistant to re-configure parameter-driven systems

Parameter-Driven Systems (PDS) are widely used in commerce for large-scale applications. Reusability is achieved with a PDS design by relocating implicit control structures in the software and the storage of explicit data in database files. This approach can accommodate various user requirements without tedious modification of the software. In order to specify appropriate parameters in a system, knowledge of both business activities and system behaviour are required. For large, complex software packages, this task becomes time consuming and requires specialist knowledge, yet the consistency and correctness still cannot be guaranteed. My research studied the types of knowledge required and agents involved in the PDS customisation. The work also identified the associated problems and constraints. A solution is proposed and implemented as an Intelligent Assistant prototype than a manual approach. Three areas of achievement have been highlighted: 1. The characteristics and problems of maintaining parameter instances in a PDS are defined. It is found that the verification is not complete with the technical/structural knowledge alone, but a context is necessary to provide semantic information and related business activities (thus the implemented parameters) so that mainline functions can relate with each other. 2. A knowledge-based modelling approach has been proposed and demonstrated via a practical implementation. A Specification Language was designed which can model various types of knowledge in a PDS and encapsulate relationships. The Knowledge-Based System (KBS) developed verifies parameters based on the interpreted model of a given context. 3. The performance of the Intelligent Assistant prototype was well received by the domain specialist from the participating organisation. The modelling and KBS approach developed in my research offers considerable promise in solving practical problems in the software industry.

Why do Australians, New Zealanders and Americans travel to Fiji? Evidence from cointegration, unit roots and parameter stability tests

The aim of this study was to estimate the demand for Fiji’s tourism from its three main source markets—Australia, New Zealand, and the US—using the bounds testing approach to cointegration. Our main finding was that visitor arrivals to Fiji and its key determinants are cointegrated over the 1970–2000 period. We then used the autoregressive distributed lag model to estimate short-run and long-run elasticities and found that income in origin countries, transport costs, and prices were significant determinants of Fiji’s tourism demand. We also found that coups negatively impact visitor arrivals from all markets. In testing for parameter stability, we established that the series were integrated of order one in the presence of a structural break. We then used the Hansen test for parameter stability and found that the parameters of our long-run model are stable over time.

Determining RF MEMS switch parameter by neural networks

A challenge in designing a RF MEMS switch is the determination of its parameters to satisfy the application requirements. Often this is done through a set of comprehensive time consuming simulations. This paper employs neural networks and develops a supervised learner that is capable of determining S11 parameter for a RF MEMS shunt switch. The inputs are the length its L and the height of its gap. The outputs are S11s for eight different frequency points from 0 to V band. The developed learner helps prevent repetitive simulations when designing the specified switch. Simulation results are presented.

Parameter sensitivity analysis of surface plasmon resonance biosensor through numerical simulation

Surface based analytical tools have gained more importance for rapid, sensitive and label-free monitoring of molecular recognition events. Surface plasmon resonance (SPR) has played a prominent role in real time monitoring of surface binding events. SPR is increasing its significance especially for the study of ultrathin dielectric layer. This paper investigates the role of thin films of gold, silver and aluminium for protein detection in SPR biosensors. It is shown that the sensitivity, which is indicated by the shift of plasmon dip, is not linearly related to the thickness of protein but quadratic over a specific range. The approach involves a plot of a reflectivity curve as a function of the angle of incidence.

Parameter-dependent H∞ control for time varying delay polytopic systems

This paper addresses the robust stabilization and H_∞ control problem for a class of linear polytopic systems with continuously distributed delays. The control objective is to design a robust H_∞ controller that satisfies some exponential stability constraints on the closed-loop poles. Using improved parameter-dependent Lyapunov Krasovskii functionals, new delay-dependent conditions for the robust H_∞ control are established in terms of linear matrix inequalities.

Automatic parameter selection for Eigenfaces

In this paper, we investigate the parameter selection issues for Eigenfaces. Our focus is on the eigenvectors and threshold selection issues. We propose a systematic approach in selecting the eigenvectors based on the relative errors of the eigenvalues. In addition, we have designed a method for selecting the classification threshold that utilizes the information obtained from the training database effectively. Experimentation was conducted on the ORL and AMP face databases with results indicating that the automatic eigenvectors and threshold selection methods provide an optimum recognition in terms of precision and recall rates. Furthermore, we show that the eigenvector selection method outperforms energy and stretching dimension methods in terms of selected number of eigenvectors and computation cost.