Search

Credit scoring is a crucial task within risk management for any company in the financial sector. On the one hand, it is in the self-interest of banks to avoid approving credits to customers who probably default. On the other hand, regulators require strict risk management systems from banks to protect their customers and, from “too big to fail institutions”, to avoid bankruptcy with negative impacts on an economy as a whole. However, credit scoring is also expensive and time-consuming. So, any possible method, like three-way decisions, to further increase its efficiency, is worth a try. We propose a two-step approach based on three-way decisions. Customers whose credit applications can be approved or rejected right away are decided in a first step. For the remaining credit applications, additional information is gathered in a second step. Hence, these decisions are more expensive than the ones in the first step. In our paper, we present a methodology to apply three-way decisions with probabilistic rough sets for credit scoring and an extensive case study with more than 7000 credit applications from Chilean micro-enterprises.

The atomic structural arrangements of liquid iron oxides affect the thermophysical and thermodynamic properties associated with the steelmaking process and magma flows. Here, the structures of stable and supercooled iron oxide melts have been investigated as a function of oxygen fugacity and temperature, using x-ray diffraction and aerodynamic levitation with laser heating. Total x-ray structure factors and their corresponding pair distribution functions were measured for temperatures ranging from 1973 K in the stable melt, to 1573 K in the deeply supercooled liquid region, over a wide range of oxygen partial pressures. Empirical potential structure refinement yields average Fe–O coordination numbers ranging from ~4.5 to ~5 over the region FeO to Fe2O3, significantly lower than most existing reports. Ferric iron is dominated by FeO4, FeO5 and FeO6 units in the oxygen rich melt. For ferrous iron under reducing conditions FeO4 and FeO5 units dominate, in stark contrast to crystalline FeO. Knowing the atomic structure of liquid iron oxide is key to understanding steelmaking processes and magma flows. Here, Fe-O atomic coordination numbers are determined during levitation melting at a range of temperatures and oxygen partial pressures, revealing low coordination numbers.

Clustering is one of the main data mining tasks with many proven techniques and successful real-world applications. However, in changing environments, the existing systems need to be regularly updated in order to describe in the best possible way an observed phenomenon at each point in time. Since changes lead to uncertainty, the respective systems also require an adequate modeling of the involved kinds of uncertainty. This paper presents a novel method for dynamic clustering called dynamic rough-fuzzy support vector clustering (D-RFSVC). Its main idea is to take advantage of the knowledge acquired in previous cycles to speed up model updating while tracking the structural changes that clusters can experience over time. The core method of the proposed approach is the well-known support vector clustering algorithm, which can be used for large datasets employing powerful optimization techniques. The computational experiments, together with a conceptual and numerical comparative study, highlight the potential D-RFSVC has in dynamic environments.

© 2018 Elsevier Ltd Transportation Research (TR) was established in 1967 with the vision of promoting multi-disciplinary (economics, engineering, sociology, psychology) research on transport systems. The journal has continuously expanded its wings becoming a world-leading journal, now publishing research work through six parts, A to F, respectively addressing Policy and Practice, Methodological, Emerging Technologies, Transport and Environment, Logistics and Transportation Review, and Traffic Psychology and Behaviour. This study aims to celebrate the first half century of the journal through a bibliometric study of the publications on all six parts between 1967 and 2016. It uses the most reliable database for academic research, the Web of Science Core Collection, to identify the leading trends in all TR journals in terms of impact, topics, authors, universities, and countries. Moreover, it uses the Visualization of Similarities (VOS) viewer software to analyse bibliographic coupling, co-citation, citation, co-authorship, and co-occurrence of keywords.

Large access providers (ISPs) are seeking for new types of business agreements and pricing models to manage network costs and monetize better the provision of last-mile services. A typical paradigm of such new pricing norms is the proliferation of paid peering deals between ISPs and content providers (CPs), while on top of this, some ISPs are already experimenting with usage-based tariffs, usually through data-plans, instead of the typical fixed-based charging. In this work we define as common platform, the infrastructure in which a single ISP transacts with several CPs through peering agreements. In this context, we examine whether, and under which market conditions, the profitability of the involved stakeholders improves when the establishment of this platform is accompanied by a monetary compensation from the CPs to the ISP (paid peering), v.s. a scenario where their deal is a typical settlement-free one. In both cases, we assume that the ISP implements a usage-based access pricing scheme, implying that end-users will pay more for higher transaction rates with the CPs. Our framework captures some of the most important details of the current market, such as the various business models adopted by the CPs, the end-users' evaluation towards the ISP's and CPs' level of investments and the traffic rates per transaction for the offered services. By analysing the equilibrium derived by a leader-follower game, it turns out (among other practical takeaways) that whether or not the profitability of a CP improves, it highly depends on whether its business model is to sell content, or if it obtains its revenue from advertisements. Finally, we extract that consumer surplus is considerably higher under paid peering, which in turn implies improved levels of social welfare1.

We present a novel soft clustering approach based on Support Vector Clustering.Data points outside the clusters found form a fuzzy boundary region.Clusters with any shape as well as outliers can be identified.Membership degrees were calculated in a natural way. Support Vector Clustering (SVC) is an important density-based clustering algorithm which can be applied in many real world applications given its ability to handle arbitrary cluster silhouettes and detect the number of classes without any prior knowledge. However, if outliers are present in the data, the algorithm leaves them unclassified, assigning a zero membership degree which leads to all these objects being treated in the same way, thus losing important information about the data set. In order to overcome these limitations, we present a novel extension of this clustering algorithm, called Rough-Fuzzy Support Vector Clustering (RFSVC), that obtains rough-fuzzy clusters using the support vectors as cluster representatives. The cluster structure is characterized by two main components: a lower approximation, and a fuzzy boundary. The membership degrees of the elements in the fuzzy boundary are calculated based on their closeness to the support vectors that represent a specific cluster, while the lower approximation is built by the data points which lie inside the hyper-sphere obtained in the training phase of the SVC algorithm. Our computational experiments verify the strength of the proposed approach compared to alternative soft clustering techniques, showing its potential for detecting outliers and computing membership degrees for clusters with any silhouette.

In adversarial classification, the interaction between classifiers and adversaries can be modeled as a game between two players. It is natural to model this interaction as a dynamic game of incomplete information, since the classifier does not know the exact intentions of the different types of adversaries (senders). For these games, equilibrium strategies can be approximated and used as input for classification models. In this paper we show how to model such interactions between players, as well as give directions on how to approximate their mixed strategies. We propose perceptron-like machine learning approximations as well as novel Adversary-Aware Online Support Vector Machines. Results in a real-world adversarial environment show that our approach is competitive with benchmark online learning algorithms, and provides important insights into the complex relations among players.

One of the most common methods used in the social network analysis of criminal groups is node importance evaluation, which focuses on the links between network members to identify likely crime suspects. Because such traditional node evaluators do not take full advantage of group members' individual criminal propensities, a new evaluator called the social network criminal suspect evaluator (SNCSE) is proposed. SNCSE incorporates members' individual criminal propensities into the node importance evaluation and employs a novel perspective based on concepts of human and social capital, an ego network structure, and an analogy between social interaction and field theory. SNCSE is applied to solve two real-world problems. Its effectiveness is compared with that of traditional evaluators. The results show that integrating criminal propensity into network analysis enables the more accurate identification of key suspects compared to alternative evaluators.

Data analysis has gained strategic importance for virtually any organization. It covers areas like business analytics, big data, business intelligence, and data mining, among others. The past decades have also witnessed increasing efforts to capture, analyze, and interpret dynamic data instead of just static snapshot data. This is due to the fact that many real-life applications are characterized by changing data structures. Hence, analytic systems need to be able to adapt to changes. In recent years, many models for dynamic data analysis have been proposed and successfully applied in a diverse range of real-life projects. Since the number of respective algorithms has continuously risen, it has become increasingly demanding to keep track in this field. This is not only related to the algorithms that have been proposed so far and their relationships to each other. It also applies to the disclosure of gaps in research that need to be filled by appropriate new algorithms and, therefore, uncover new research opportunities. To contribute to the review of this field, we propose a holistic framework to categorize dynamic data analysis, the dynXcube-Framework. We show that dynXcube is very useful to present the state-of-the-art of dynamic data analysis in a consolidated way. Furthermore, it has the potential to disclose gaps in current research, thus providing a road map for future activities in the field of dynamic data analysis. Therefore, dynXcube is a significant step towards an improved accessibility of dynamic data analysis methods for academics and professionals alike and will help to stimulate future research in this important field.

Churn prediction is an important application of classification models that identify those customers most likely to attrite based on their respective characteristics described by e.g. socio-demographic and behavioral variables. Since nowadays more and more of such features are captured and stored in the respective computational systems, an appropriate handling of the resulting information overload becomes a highly relevant issue when it comes to build customer retention systems based on churn prediction models. As a consequence, feature selection is an important step of the respective classifier construction process. Most feature selection techniques; however, are based on statistically inspired validation criteria, which not necessarily lead to models that optimize goals specified by the respective organization. In this paper we propose a profit-driven approach for classifier construction and simultaneous variable selection based on Support Vector Machines. Experimental results show that our models outperform conventional techniques for feature selection achieving superior performance with respect to business-related goals. publisher: Elsevier articletitle: Profit-based feature selection using support vector machines – General framework and an application for customer retention journaltitle: Applied Soft Computing articlelink: http://dx.doi.org/10.1016/j.asoc.2015.05.058 content_type: article copyright: Copyright © 2015 Elsevier B.V. All rights reserved. ispartof: Applied Soft Computing vol:35 pages:740-748 status: published

This special issue on “Data Analysis and Intelligent Optimization with Applications” follows a previous special issue of this journal on the interplay of Machine Learning and Optimization, “Model Selection and Optimization in ML” (Machine Learning 85:1-2, October 2011). This time we shift our focus to applications of data analysis and optimization techniques. Optimization problems underliemostmachine learning approaches.Due to emergence of new practical applications, new problems and challenges for traditional approaches arise. Emergent applications generate new data analysis problems, which, in turn boost new research in optimization. The contribution of machine learning researchers into the field of optimization is of considerable significance and should not be overlooked. This special issue collected solutions, adapted for real world problems, leading to massive and large-scale data sets, online data and imbalanced data. We encouraged submission of papers, devoted to combining machine learning and data analysis techniques with advances in optimization to produce methods of Intelligent Optimization, both theoretical and practical. Our goal for this special issue was to bring together researchers working in different areas, related to analytics and optimization. The special issue includes papers on a variety of topics, including, but not limited to

One of the main tasks of conjoint analysis is to identify consumer preferences about potential products or services. Accordingly, different estimation methods have been proposed to determine the corresponding relevant attributes. Most of these approaches rely on the post-processing of the estimated preferences to establish the importance of such variables. This paper presents new techniques that simultaneously identify consumer preferences and the most relevant attributes. The proposed approaches have two appealing characteristics. Firstly, they are grounded on a support vector machine formulation that has proved important predictive ability in operations management and marketing contexts and secondly they obtain a more parsimonious representation of consumer preferences than traditional models. We report the results of an extensive simulation study that shows that unlike existing methods, our approach can accurately recover the model parameters as well as the relevant attributes. Additionally, we use two conjoint choice experiments whose results show that the proposed techniques have better fit and predictive accuracy than traditional methods and that they additionally provide an improved understanding of customer preferences.

This paper describes the direct measurements of the structure of iron-bearing liquids using a combination of containerless techniques and in-situ high energy x-ray diffraction.These capabilities provide data that is important to help model and optimize processes such as smelting, steel making, and controlling slag chemistry. A successful programme of liquid studies has been undertaken and the Advanced Photon Source using these combined techniques which include the provision of gas mixing and the control of pO2and the changing influence of mixed valance elements. It is possible to combine rapid image acquisition with quenching of liquids to obtain the full diffraction patterns of deeply supercooled liquids and the metastable supercooled liquid regime, where the liquid structures and viscosity change most dramatically, can also be explored.

Feature selection and classification of imbalanced data sets are two of the most interesting machine learning challenges, attracting a growing attention from both, industry and academia. Feature selection addresses the dimensionality reduction problem by determining a subset of available features to build a good model for classification or prediction, while the class-imbalance problem arises when the class distribution is too skewed. Both issues have been independently studied in the literature, and a plethora of methods to address high dimensionality as well as class-imbalance has been proposed. The aim of this work is to simultaneously explore both issues, proposing a family of methods that select those attributes that are relevant for the identification of the target class in binary classification. We propose a backward elimination approach based on successive holdout steps, whose contribution measure is based on a balanced loss function obtained on an independent subset. Our experiments are based on six highly imbalanced microarray data sets, comparing our methods with well-known feature selection techniques, and obtaining a better prediction with consistently fewer relevant features.

The long-standing network neutrality debate concerns the future of pricing practices in the Internet [2, 5]. Presently, an Internet service provider (ISP) charges both its enduser subscribers and its directly connected content service providers (CSPs) for access to its network and the whole Internet. A central question is whether this practice should continue, or whether each ISP should be permitted to impose charges upon all content providers that deliver content to the ISP’s subscribers (not only upon those that attach to the ISP directly). The debate is thus between the existing practice of ‘one-sided pricing’ (also called, ‘neutral’ pricing), in which the ISP obtains all its revenue by only charging its end-users, and ‘two-sided pricing’ (non-neutral) [1, 9]. There is a substantial literature on the subject, [3, 4, 6, 7], focusing on the welfare effects of two-sided pricing compared to neutral pricing. Many different models have been proposed; using non-comparable sets of parameters and assumptions, they have produced differing answers. In this paper we present a model which is both simple and general, based on the model proposed in [8]. It captures the essential aspects of investment incentives by ISPs and CSPs and how these are influenced by the different ways that ISPs might charge. We allow for the fact that CSPs may be of different sizes and that they can obtain revenue from selling content to end-users (as well as from from advertisement, as in [8]). We simplify the model by focusing on a single ISP who interacts with many independent CSPs. As in [8] we analyse what happens at the symmetric Nash equilibrium of the induced leader-follower game. Our results are not concerned with the welfare effects of adopting alternative pricing policies, but on the incentives of the CSPs to accept them. We are able to make some interesting observations about the incentives provided to large and small CSPs. If all CSPs have as their business model to sell content to end-users at some strategically defined price, then both large and small CSPs will have aligned incentives towards adopting or refusing a two-sided pricing policy if this is proposed by the ISP; it is either beneficial for all or for none. In implementing two-sided pricing, an ISP might use either a ‘uniform’ or ‘differentiated’ pricing strategy; in the latter case the ISP charges each CSP a price that depends on the CSP’s size, and in the former case not. CSPs may have differing preferences for these strategies. On the one hand, if all CSPs aim

Policing plays an important role in combating street crime. Though policing actions have a dissuasive impact on criminal behavior, they can also have unpredictable and even undesirable effects such as displacement of crime hot-spots. This paper presents an agent-based simulation model that generates artificial street-crime data which can be used to test different policing strategies in a virtual environment. The model can thus evaluate the strategies’ effectiveness and collateral effects before putting them into practice and provide support for the policing decision-making process. Based on this model, a crime simulator was implemented in Repast Simphony and a series of test simulations on fictitious and real cities were carried out. The proposed formulation was successfully validated, confirming its potential as a powerful tool for the study of street crime.

Today a spectrum of solutions are available for istributing content over the Internet, ranging from commercial CDNs to ISP-operated CDNs to content-provider-operated CDNs to peer-to-peer CDNs. Some deploy servers in just a few large data centers while others deploy in thousands of locations or even on millions of desktops. Recently, major CDNs have formed strategic alliances with large ISPs to provide content delivery network solutions. Such alliances show the natural evolution of content delivery today driven by the need to address scalability issues and to take advantage of new technology and business opportunities. In this paper we revisit the design and operating space of CDN-ISP collaboration in light of recent ISP and CDN alliances. We identify two key enablers for supporting collaboration and improving content delivery performance: informed end-user to server assignment and in-network server allocation. We report on the design and evaluation of a prototype system, NetPaaS, that materializes them. Relying on traces from the largest commercial CDN and a large tier-1 ISP, we show that NetPaaS is able to increase CDN capacity on-demand, enable coordination, reduce download time, and achieve multiple traffic engineering goals leading to a win-win situation for both ISP and CDN.

In recent years, wireless sensor networks became popular for a wide range of mainstream applications. Closely related with this evolution, a problem for consumer market use emerged: How to initialize and setup the infrastructure automatically. This paper presents an approach to solve this problem. We present a novel approach how to build infrastructure maps only with anchor-mobile range measurements. The approach uses a baseline SLAM implementation in form of incremental posterior mapping. We adapt the approach by representing mobile posterior as well as anchor maps with probability grids similar to Markov Localization due to addressing the complex Range Only Simultaneous Localization and Mapping (RO-SLAM) problem. In urban areas mobiles are employed e.g. by pedestrians or bikes which feature a specific kinematic locomotion activity. Hence, we pair RO-SLAM with a SVM-based activity classifier in order to raise anchor mapping accuracy. Simulation results discuss algorithm convergence and demonstrate the accuracy improvement in the presence of activity information.

Paid peering is controversial. Generally speaking, large access ISPs favour it, while Content Providers (CPs) do not. But is CP opposition to paid peering irrational, especially if their transferred payments can incentivize the ISP to make greater investment in the common infrastructure? To answer this question we analyze an ecosystem consisting of a single access ISP that directly interconnects with multiple CPs, doing so in two extreme situations: when paid peering is used for all the involved CPs, and when all parties agree on a settlement-free deal. We analyse the equilibrium, finding the total amount of investments and profits of the various stakeholders, when their revenues are affected by the total investment in infrastructure and the charges that result from the different agreements. Interestingly, it turns out whether or not there is benefit to a CP, depends on whether its business model is to sell content, and so it can recover part of the paid peering charge by raising prices to its customers, or if the CP obtains its revenue from ads. It also depends on the volume (in traffic units) per customer transaction, since we find that CPs with high volume per transaction will be charged less per byte, and so will contribute proportionally less to the total paid peering revenue that subsidizes the common ISP infrastructure. We also find a crucial role for the end-users' evaluation of the level of investments of the various market players. Another relevant factor is the access price that an ISP charges its customers.

We introduce an embedded method that simultaneously selects relevant features during classifier construction by penalizing each feature's use in the dual formulation of support vector machines (SVM). This approach called kernel-penalized SVM (KP-SVM) optimizes the shape of an anisotropic RBF Kernel eliminating features that have low relevance for the classifier. Additionally, KP-SVM employs an explicit stopping condition, avoiding the elimination of features that would negatively affect the classifier's performance. We performed experiments on four real-world benchmark problems comparing our approach with well-known feature selection techniques. KP-SVM outperformed the alternative approaches and determined consistently fewer relevant features.

In adversarial systems, the performance of a classifier decreases after it is deployed, as the adversary learns to defeat it. Recently, adversarial data mining was introduced as a solution to this, where the classification problem is viewed as a game mechanism between an adversary and an intelligent and adaptive classifier. Over the last years, phishing fraud through malicious email messages has been a serious threat that affects global security and economy, where traditional spam filtering techniques have shown to be ineffective. In this domain, using dynamic games of incomplete information, a game theoretic data mining framework is proposed in order to build an adversary aware classifier for phishing fraud detection. To build the classifier, an online version of the Weighted Margin Support Vector Machines with a game theoretic prior knowledge function is proposed. In this paper, a new content-based feature extraction technique for phishing filtering is described. Experiments show that the proposed classifier is highly competitive compared with previously proposed online classification algorithms in this adversarial environment, and promising results where obtained using traditional machine learning techniques over extracted features.

Traditional methodologies for time series prediction take the series to be predicted and split it into training, validation, and test sets. The first one serves to construct forecasting models, the second set for model selection, and the third one is used to evaluate the final model. Different time series approaches such as ARIMA and exponential smoothing, as well as regression techniques such as neural networks and support vector regression, have been successfully used to develop forecasting models. A problem that has not yet received proper attention, however, is how to update such forecasting models when new data arrives, i.e. when a new event of the considered time series occurs. This paper presents a strategy to update support vector regression based forecasting models for time series with seasonal patterns. The basic idea of this updating strategy is to add the most recent data to the training set every time a predefined number of observations takes place. This way, information in new data is taken into account in model construction. The proposed strategy outperforms the respective static version in almost all time series studied in this work, considering three different error measures.

We introduce a novel wrapper Algorithm for Feature Selection, using Support Vector Machines with kernel functions. Our method is based on a sequential backward selection, using the number of errors in a validation subset as the measure to decide which feature to remove in each iteration. We compare our approach with other algorithms like a filter method or Recursive Feature Elimination SVM to demonstrate its effectiveness and efficiency.

I. Leonard Bernstein, MD; James T. Li, MD, PhD; David I. Bernstein, MD; Robert Hamilton, PhD, DABMLI; Sheldon L. Spector, MD; Ricardo Tan, MD; Scott Sicherer, MD; David B. K. Golden, MD; David A. Khan, MD; Richard A. Nicklas, MD; Jay M. Portnoy, MD; Joann Blessing-Moore, MD; Linda Cox, MD; David M. Lang, MD; John Oppenheimer, MD; Christopher C. Randolph, MD; Diane E. Schuller, MD; Stephen A. Tilles, MD; Dana V. Wallace, MD; Estelle Levetin, PhD; and Richard Weber, MD

We present Rough-Fuzzy Support Vector Domain Description (RFSVDD), a novel data description algorithm that provides a rough-fuzzy characterization of a data set and shows its potential for outlier detection. Its resulting data structure is characterized by two components: a crisp lower-approximation and a fuzzy boundary. While the lower-approximation consists of those data points that lie inside the hypersphere, obtained by traditional Support Vector Domain Description (SVDD), the membership degree of each element in the fuzzy boundary is calculated based on its closeness to the support vectors that define the frontier of the lower-approximation. This implies a quite natural way of determining membership values without assuming any prior knowledge about the data. Our computational experiments emphasize the strength of the proposed approach underlining its potential for outlier detection.

Localization methods play an increasing role in many of today's Intelligent Transportation Systems (ITS). In the context of this work we concentrate on improving the localization accuracy of passengers in public transport vehicles, e.g. as part of automated registration, counting or ticketing systems. In these applications each passenger's ticket could be designed as a mobile wireless sensor node, which is registered to be inside or outside the vehicle using a network of fixed anchor nodes inside only. We shortly describe two specific positioning approaches (SMDS, GBP) out of a variety of methods. A set of measurements conducted in a realistic bus cabin environment is then used to analyze the performance of both algorithms. The accuracy of the proposed GBP-estimator is compared with a previously studied SMDS-estimator, particularly in severe multipath environments like in the given passenger cabin.

Novel embedded feature selection approach for Kernel K-means.Optimization procedure based on scaling factors and a concave penalization.Best performance in terms of clustering quality and dimensionality reduction. We present an unsupervised method that selects the most relevant features using an embedded strategy while maintaining the cluster structure found with the initial feature set. It is based on the idea of simultaneously minimizing the violation of the initial cluster structure and penalizing the use of features via scaling factors. As the base method we use Kernel K-means which works similarly to K-means, one of the most popular clustering algorithms, but it provides more flexibility due to the use of kernel functions for distance calculation, thus allowing the detection of more complex cluster structures. We present an algorithm to solve the respective minimization problem iteratively, and perform experiments with several data sets demonstrating the superior performance of the proposed method compared to alternative approaches.

Various techniques have been proposed to forecast a given time series. Models from the ARIMA family have been successfully used, as well as regression approaches based on e.g. linear, non-linear regression, neural networks, and Support Vector Regression. What makes the difference in many real-world applications, however, is not the technique but an appropriate forecasting methodology. Here, we propose such a methodology for the regression-based forecasting approach. A hybrid system is presented that iteratively selects the most relevant features and constructs the regression model optimizing its parameters dynamically. We develop a particular technique for feature selection as well as for model construction. The methodology, however, is a generic one providing the opportunity to employ alternative approaches within our framework. The application to several time series underlines its usefulness.

The Internet has become an important medium for effective marketing and efficient operations for many institutions. Visitors of a particular web site leave behind valuable information on their preferences, requirements, and demands regarding the offered products and/or services. Understanding these requirements online, i.e., during a particular visit, is both a difficult technical challenge and a tremendous business opportunity. Web sites that can provide effective online navigation suggestions to their visitors can exploit the potential inherent in the data such visits generate every day. However, identifying, collecting, and maintaining the necessary knowledge that navigation suggestions are based on is far from trivial. We propose a methodology for acquiring and maintaining this knowledge efficiently using data mart and web mining technology. Its effectiveness has been shown in an application for a bank's web site.

Dynamic data mining is increasingly attracting attention from the respective research community. On the other hand, users of installed data mining systems are also interested in the related techniques and will be even more since most of these installations will need to be updated in the future. For each data mining technique used, we need different methodologies for dynamic data mining. In this paper, we present a methodology for dynamic data mining based on fuzzy clustering. Using the implementation of the proposed system we show its benefits in two application areas: customer segmentation and traffic management.

Web usage mining is an important and fast developing area of web mining where a lot of research has been done already. Recently, companies got aware of its potentials, especially for applications in marketing. A structured methodology is, however, a crucial requirement for a successful practical application of web usage mining. This publication provides such a methodology that is based on suggestions from literature and own experience from various web mining projects. Its application in a Chilean bank shows how a combined use of data from a data warehouse and web data can contribute to improve marketing activities. The benefits from this project point at the huge potential web usage mining has not only in financial services.

Glasses based on rare earth oxides and aluminum oxide with 0–20 mol% SiO 2 provide a combination of optical properties, mechanical and chemical stability, and process characteristics not available in other oxide materials. Properties of the glasses include: refractive index 1.7–1.8, low dispersion (Abbe number ∼40), high solubility of optically active dopants, homogeneous chemical composition, long fluorescence lifetimes at dopant concentrations up to 5 mol%, broad fluorescence bandwidth, and infrared transmission to ≈5000 nm. The glasses are hard, strong and resist chemical attack and they can be cast in sections 5–10 mm thick. This paper briefly describes glass processing and presents bulk glass properties including results of experiments to study infrared fluorescence at wavelengths ∼1550, ∼2900 and ∼1030 nm from Er- and Yb-doped glasses that were optically pumped at 980 nm.

Determining what fertilizer mix to apply to certain soils is complicated and time consuming. Experts vaguely describe the complex relationships among the influencing parameters, such as fertilizer, soil, weather characteristics, and the previous and target yields. Farmers expect fast and reliable recommendations for fertilizer mixes that will meet their needs at low prices. Like buyers of many products, they take the quality of the products for granted. What differentiates providers is the service they offer their customers. Recognizing this fact, Soquimich S.A., a fertilizer producer, initiated development of a system that would automatically determine optimal fertilizer blends. We developed a hybrid tool using expert system technology and mixed-integer linear-programming models that increased fertilizer sales. Our initial highly sophisticated preliminary version of the system proved unsatisfactory because farmers could not always provide the necessary input and users did not know how to use and maintain the system. Our final simplified version of the system was successful and is being extended to other regions in Chile and in other countries.

The following article discusses the current emphasis and attention being given to the future of emergency management, as well as theoretical constructs designed to guide research and help practitioners reduce disaster. It illustrates that while the disaster-resistant community, disaster-resilient community, and sustainable development/sustainable hazards mitigation concepts provide many unique advantages for disaster scholarship and management, they fail to sufficiently address the triggering agents, functional areas, actors, variables, and disciplines pertaining to calamitous events. In making this argument, the article asserts that any future paradigm and policy guide must be built on—yet go further than—comprehensive emergency management. The article also reviews and alters the concept of invulnerable development. Finally, the article presents “comprehensive vulnerability management” as a paradigm and suggests that it is better suited to guide scholarly and practitioner efforts to understand and reduce disasters than the aforementioned perspectives.

The performance of classification methods, such as Support Vector Machines, depends heavily on the proper choice of the feature set used to construct the classifier. Feature selection is an NP-hard problem that has been studied extensively in the literature. Most strategies propose the elimination of features independently of classifier construction by exploiting statistical properties of each of the variables, or via greedy search. All such strategies are heuristic by nature. In this work we propose two different Mixed Integer Linear Programming formulations based on extensions of Support Vector Machines to overcome these shortcomings. The proposed approaches perform variable selection simultaneously with classifier construction using optimization models. We ran experiments on real-world benchmark datasets, comparing our approaches with well-known feature selection techniques and obtained better predictions with consistently fewer relevant features.

Operators of multi?service networks require simple charging schemes with which they can fairly recover costs from their users and effectively allocate network resources. This paper studies an approach for computing such charges from simple measurements (time and volume), and relating these to bounds of the effective bandwidth. To achieve economic efficiency, it is necessary that usage?based charging schemes capture the relative amount of resources used by connections. Based on this criteria, we evaluate our approach for real traffic consisting of Internet Wide Area Network traces and MPEG?1 compressed video. Its incentive compatibility is shown with an example involving deterministic multiplexing, and the effect of pricing on a network's equilibrium is investigated for deterministic and statistical multiplexing. Finally, we investigate the incentives for traffic shaping provided by the approach.

The main objective of machine diagnosis is the early recognition of mechanical defects in a machine, which is often referred to as preventive maintenance. This may result in the reduction of faults and higher machine availability. Preventive maintenance can be performed periodically in fixed time intervals, by demand due to machine faults, or continuously, depending on the state of a machine. To minimize maintenance and repair time, state-dependent maintenance of machines is usually applied. It assumes precise and reliable monitoring of machine's states, which is often provided by an expert who can distinguish malfunctions in operation from other changes in machine's states. In Fuzzy-Neuro Systems '98 — Computational Intelligence, pp. 98–105, and Int. J. Fuzzy Sets and Systems 105 (1999) 81–90, we have introduced a clustering method for dynamic objects, i.e., objects which are described by (time) trajectories of features. The aim of this paper is to show exemplarily that this method can be used for early recognition of changes in a machine's state and thus for automatic fault detection. Moreover, the method can be applied to automatic feature selection. A major advantage of this method is that it requires less expert knowledge than traditional approaches.

In recent years, overlay networks have proven a popular way of disseminating potentially large files from a single server S to a potentially large group of N end users via the Internet. A number of algorithms and protocols have been suggested, implemented and studied. In particular, much attention has been given to peer-to-peer (P2P) systems such as BitTorrent, Slurpie, SplitStream, Bullet and Avalanche. The key idea is that the file is divided into M parts of equal size and that a given user may download any one of these -- or, for Avalanche, linear combinations of these -- either from the server or from a peer who has previously downloaded it. However, performance analysis of P2P systems for file dissemination has typically been limited to comparing one system relative to another and typically been realized by means of simulations and measurements. We give the minimal time to fully disseminate the file of M parts from a server to N end users in a centralized scenario. In the scheduling literature this completion time is referred to as makespan. We thereby provide a lower bound which can be used as a performance benchmark for any P2P file dissemination system. We also investigate the part of the loss in efficiency that is due to the lack of centralized control in practice. Using simulation as well as direct computation, we show that even a simple and natural randomized strategy disseminates the file in an expected time that grows with N in a similar manner to the minimal time achieved with a centralized controller. This suggests that the performance of necessarily decentralized P2P file dissemination systems should still be close to our performance bound.

A new levitation apparatus coupled to a synchrotron-derived x-ray source has been developed to study the structure of liquids at temperatures up to 3000 K. The levitation apparatus employs conical nozzle levitation using aerodynamic forces to stably position solid and liquid specimens at high temperatures. A 270 W CO2 laser was used to heat the specimens to desired temperatures. Two optical pyrometers were used to record the specimen temperature, heating curves, and cooling curves. Three video cameras and a video recorder were employed to obtain and record specimen views in all three dimensions. The levitation assembly was supported on a three-axis translation stage to facilitate precise positioning of the specimen in the synchrotron radiation beam. The levitation system was enclosed in a vacuum chamber with Be windows, connections for vacuum and gas flow, ports for pyrometry, video, and pressure measurements. The vacuum system included automatic pressure control and multi-channel gas flow control. A phos...

This paper first provides an overview of data preprocessing, focusing on problems of real world data. These are primarily problems that have to be carefully understood and solved before any data analysis process can start. The paper discusses in detail two main reasons for performing data preprocessing: i problems with the data and ii preparation for data analysis. The paper continues with details of data preprocessing techniques achieving each of the above mentioned objectives. A total of 14 techniques are discussed. Two examples of data preprocessing applications from two of the most data rich domains are given at the end. The applications are related to semiconductor manufacturing and aerospace domains where large amounts of data are available, and they are fairly reliable. Future directions and some challenges are discussed at the end.

In this work we use wireless sensor networks (WSNs) to estimate positions for automated passenger registration with the purpose of an electronic ticketing application in public transport systems. The network is capable of measuring distances between the network's sensor nodes by phase-of-arrival (POA) ranging. In our application, the requirements regarding position accuracy are very demanding, due to the ranging biases caused by severe non-line-of-sight (NLOS) environments, i.e. highly reflective radio propagation. Hence, we suggest semidefinite programming (SDP) as a robust optimization algorithm for the localization process and apply a NLOS mitigation technique from literature to a modified cost function that may help to meet these requirements. Finally the proposed algorithm is compared with state-of-the-art positioning techniques. Testing with simulated and real measured data shows that the proposed SDP estimator outperforms the other positioning algorithms substantially, particularly in realistic NLOS environments.

Localization techniques based on wireless sensor networks (WSNs) are an increasingly popular approach for estimating object positions in a wide area of applications. Nevertheless, accuracy and reliability of the WSN position estimates need to be increased for some applications, e.g. automated people registration in public transport vehicles. This goal can be achieved by incorporating additional sensors like cameras in the localization process. In this paper we introduce a novel data fusion approach for combining WSN position estimates and object positions detected in camera images. We project image coordinates to the WSN coordinate frame and modify an Extended Kalman Filter (EKF) for data fusion. We show how the object positions that arise from 2D camera images are used to reduce the variance of the combined position estimate. We test our method by tracking the movement of a person using WSN positioning and additional measurements obtained by people detection in the corresponding video scene.

Addresses the issue of call acceptance and routing in ATM networks. The goal is to design an algorithm that guarantees bounds on the fraction of cells lost by a call. The method proposed for call acceptance and routing does not require models describing the traffic. Each switch estimates the additional fraction of cells that would be lost if new calls were routed through the switch. The routing algorithm uses these estimates. The estimates are obtained by monitoring the switch operations and extrapolating to the situation where more calls are routed through the switch. The extrapolation is justified by a scaling property. To reduce the variance of the estimates, the switches calculate the cell loss that would occur with virtual buffers. A way to choose the sizes of the virtual buffers in order to minimize the variance is discussed. Thus, the switches constantly estimate their spare capacity. Simulations were performed using Markov fluid sources to test the validity of the approach. >

We consider localization techniques based on wireless sensor networks (WSN), which are a promising technology to achieve highly accurate positioning solutions. Hence, this paper provides an overview of the concept of localization in modern railway applications, e.g. locating trains in storage sidings. Different geometrical scenarios are analyzed by using real measurement data. Generally, position estimation in WSNs is based on distance measuring by means of time-of-flight (TOF) ranging techniques. To employ radio frequency — (RF) based positioning here, we introduce weighting combinations least squares (WECOLS), an extended LS positioning approach, for computing positions by determining the space of possible estimations generated from the input data. Furthermore we investigate WECOLS according to accuracy compared with LS and least median squares (LMS).

Although important both for educational production (e.g. Sacerdote 2001) and the formation of entrepreneurial activity (e.g. Nanda and Sorensen 2010), the effects of social interactions on the development of entrepreneurial skills in entrepreneurship education have been completely ignored so far. At the same time there is a large potential for leveraging social interaction effects to increase the formation of entrepreneurial skills among the students: team-based business planning is a prevalent component of entrepreneurship education (Krueger et al. 2000). And due to restricted education budgets, an efficient measure, for example the reshuffling of teams to maximize beneficial social interaction effects for skill formation, is welcome to instructors and course planners. A quantification of externalities deriving from social interactions would be valuable to several stakeholder groups involved in entrepreneurship education, that are concerned about efficient educational production given heterogeneous student quality (Foster 2006).

Dynamic data mining has gained increasing attention in the last decade. It addresses changing data structures which can be observed in many real-life applications, e.g. buying behavior of customers. As opposed to classical, i.e. static data mining where the challenge is to discover pattern inherent in given data sets, in dynamic data mining the challenge is to understand - and in some cases even predict - how such pattern will change over time. Since changes in general lead to uncertainty, the appropriate approaches for uncertainty modeling are needed in order to capture, model, and predict the respective phenomena considered in dynamic environments. As a consequence, the combination of dynamic data mining and soft computing is a very promising research area. The proposed algorithm consists of a dynamic clustering cycle when the data set will be refreshed from time to time. Within this cycle criteria check if the newly arrived data have structurally changed in comparison to the data already analyzed. If yes, appropriate actions are triggered, in particular an update of the initial settings of the cluster algorithm. As we will show, rough clustering offers strong tools to detect such changing data structures. To evaluate the proposed dynamic rough clustering algorithm it has been applied to synthetic as well as to real-world data sets where it provides new insights regarding the underlying dynamic phenomena.

Operations research methods and applications have played a crucial role in the development of several Chilean industries as well as the government. This paper provides a brief overview of the history and current developments of the Chilean Institute of Operations Research, ICHIO. Keywords: operations research; Chile