Ant colony and particle swarm optimization for financial classification problems

Financial decisions are often based on classification models which are used to assign a set of observations into predefined groups. Such models ought to be as accurate as possible. One important step towards the development of accurate financial classification models involves the selection of the appropriate independent variables (features) which are relevant for the problem at hand. This is known as the feature selection problem in the machine learning/data mining field. In financial decisions, feature selection is often based on the subjective judgment of the experts. Nevertheless, automated feature selection algorithms could be of great help to the decision-makers providing the means to explore efficiently the solution space. This study uses two nature-inspired methods, namely ant colony optimization and particle swarm optimization, for this problem. The modelling context is developed and the performance of the methods is tested in two financial classification tasks, involving credit risk assessment and audit qualifications.     

Collaborative recommender systems: Combining effectiveness and efficiency

Recommender systems base their operation on past user ratings over a collection of items, for instance, books, CDs, etc. Collaborative filtering (CF) is a successful recommendation technique that confronts the “information overload” problem. Memory-based algorithms recommend according to the preferences of nearest neighbors, and model-based algorithms recommend by first developing a model of user ratings. In this paper, we bring to surface factors that affect CF process in order to identify existing false beliefs. In terms of accuracy, by being able to view the “big picture”, we propose new approaches that substantially improve the performance of CF algorithms. For instance, we obtain more than 40% increase in precision in comparison to widely-used CF algorithms. In terms of efficiency, we propose a model-based approach based on latent semantic indexing (LSI), that reduces execution times at least 50% than the classic CF algorithms.     

Geometrical methods for level set based abdominal aortic aneurysm thrombus and outer wall 2D image segmentation

Abdominal aortic aneurysm (AAA) is a localized dilatation of the aortic wall. Accurate measurements of its geometric characteristics are critical for a reliable estimate of AAA rupture risk. However, current imaging modalities do not provide sufficient contrast to distinguish thrombus from surrounding tissue thus making the task of segmentation quite challenging. The main objective of this paper is to address this problem and accurately extract the thrombus and outer wall boundaries from cross sections of a 3D AAA image data set (CTA). This is achieved by new geometrical methods applied to the boundary curves obtained by a Level Set Method (LSM). Such methods address the problem of leakage of a moving front into sectors of similar intensity and that of the presence of calcifications. The versatility of the methods is tested by creating artificial images which simulate the real cases. Segmentation quality is quantified by comparing the results with a manual segmentation of the slices of ten patient data sets. Sensitivity to the parameter settings and reproducibility are analyzed. This is the first work to our knowledge that utilizes the level set framework to extract both the thrombus and external AAA wall boundaries.     

A GIS-based methodology for identifying pedestrians’ crossing patterns

A pedestrian trip is a spatiotemporal process going through different states and related to different decisions made at certain times and locations on the urban network. The analysis of pedestrian trips in terms of crossing patterns is a complex task, which is often further limited by a lack of appropriate and detailed data. The objective of this research is the development and testing of appropriate indicators of pedestrian crossing behavior along urban trips, and a methodology for collecting and processing the data required for the analysis of this behavior. First, a comprehensive set of indicators for the assessment of pedestrian behavior in urban areas is proposed (i.e. average trip length, number, type and location of crossings). Then, a GIS tool is developed for the storage and integration of information on pedestrian trips, and the crossings made during the trips, with other geographical information (e.g. road network function and geometry, traffic control and pedestrian facilities). The proposed approach is then tested at network level on a sample of pedestrian trips collected by a field survey. The results suggest specific patterns of pedestrian crossing behavior, such as the tendency to cross at the beginning of the trip and the tendency to cross at mid-block locations when signalized junctions are not available. The results are further discussed in terms of urban planning and management implications. It is concluded that the proposed approach is very efficient for the analysis of pedestrian crossing behavior in urban areas.     

Incentive Compatible Pricing Strategies for QoS Routing

QoS routing mechanisms allow users identify paths that can accommodate their performance requirements and reserve the necessary resources. An important problem is how to conduct such resource allocation efficiently, not only from the single-connection, but also from the network point of view. We propose the use of pricing mechanisms as a means to regulate the users' decisions in a networkwide efficient manner. Focusing on QoS architectures that employ rate-based schedulers, we formulate a congestion-based pricing scheme. We establish the structure of the corresponding user-optimal response, i.e., a path selection algorithm that satisfies the user's requirements at minimal cost. We show that the underlying noncooperative game among users has a unique equilibrium, for any particular choice of price functions. Then, we establish the existence of incentive compatible price functions, which drive the network into an equilibrium point that coincides with the optimum of a social function. Specifically, these price functions are the derivatives of the social function. We then extend our results to scenarios in which users can identify only sub-optimal paths, as is often the case with multi-constrained path optimization.     

Engineering the xenobiotic substrate specificity of maize glutathione S-transferase I

Glutathione S-transferases (GSTs) are a heterogeneous family of enzymes that catalyse the conjugation of glutathione (GSH) to electrophilic sites on a variety of hydrophobic substrates. In the present study three amino acid residues (Trp12, Phe35 and Ile118) of the xenobiotic binding site (H-site) of maize GST I were altered in order to evaluate their contribution to substrate binding and catalysis. These residues are not conserved and hence may affect substrate specificity and/or product dissociation. The results demonstrate that these residues are important structural moieties that modulate an enzyme's catalytic efficiency and specificity. Phe35 and Ile118 also participate in k(cat) regulation by affecting the rate-limiting step of the catalytic reaction. The effect of temperature on the catalytic activity of the wild-type and mutant enzymes was also investigated. Biphasic Arrhenius and Eyring plots for the wild-type enzyme showed an apparent transition temperature at 35 degrees C, which seems to be the result of a change in the rate-limiting step of the catalytic reaction. Thermodynamic analysis of the activity data showed that the activation energy increases at low temperatures, whereas the entropy change seems to be the main determinant that contributes to the rate-limiting step at high temperatures.     

Thermoregulatory and cardiovascular responses to creatine, glycerol and alpha lipoic acid in trained cyclists

Background

It has been shown that supplementation with creatine (Cr) and glycerol (Gly), when combined with glucose (Glu) necessary for the enhancement of Cr uptake by skeletal muscle, induces significant improvements in thermoregulatory and cardiovascular responses during exercise in the heat.

Purpose

To determine whether Cr/Gly-induced thermoregulatory and cardiovascular responses are maintained when the majority (~75%) of the Glu in the Cr/Gly supplement is replaced with the insulintropic agent alpha lipoic acid (Ala).

Methods

22 healthy endurance trained cyclists were randomly assigned to receive either 20?g/day (4?×?5?g/day) of Cr, 2?g ^.kg^-1 BM per day (4?×?0.5?g ^.kg^-1 BM per day) of Gly and 150?g/day (4?×?37.5?g/day) of Glu or 20?g/day (4?×?5?g/day) of Cr monohydrate, 2?g ^.kg^-1 BM per day (4?×?0.5?g ^.kg^-1 BM per day) of Gly (100?g/day (4?×?25?g/day) of Glu and 1000?mg/day (4?×?250?mg/day) of Ala for 7?days for 7?days. Exercise trials were conducted pre- and post-supplementation and involved 40?min of constant-load cycling exercise at 70% O₂ max by a self-paced 16.1?km time trial at 30°C and 70% relative humidity.

Results

Median and range values of TBW increased significantly by 2.1 (1.3-3.3) L and 1.8 (0.2-4.6) L in Cr/Gly/Glu and Cr/Gly/Glu/Ala groups respectively (P?=?0.03) and of BM not significantly by 1.8 (0.2-3.0) kg and 1.2 (0.5-2.1) kg in Cr/Gly/Glu and in Cr/Gly/Glu/Ala, respectively (P?=?0.75). During constant load exercise, heart rate (HR) and core temperature (Tcore) were significantly lower post-supplementation: HR was reduced on average by 3.3?±?2.1 beats/min and by 4.8?±?3.3 beats/min (mean ± SD) and Tcore by 0.2?±?0.1 (mean ± SD) in the Cr/Gly/Glu and Cr/Gly/Glu/Ala, respectively The reduction in HR and Tcore was not significantly different between the supplementation groups.

Conclusions

In comparison to the established hyper hydrating Cr/Gly/Glu supplement, supplement containing Cr/Gly/Ala and decreased amount of Glu provides equal improvements in thermoregulatory and cardiovascular responses during exercise in the heat.     

Parallel computational strategies for structural optimization

The objective of this paper is to investigate the efficiency of various computational algorithms implemented in the framework of structural optimization methods based on evolutionary algorithms. In particular, the efficiency of parallel computational strategies is examined with reference to evolution strategies (ES) and genetic algorithms (GA). Parallel strategies are implemented both at the level of the optimization algorithm, by exploiting the natural parallelization features of the evolutionary algorithms, as well as at the level of the repeated structural analysis problems that are required by ES and GA. In the latter case the finite element solutions are performed by the FETI domain decomposition method specially tailored to the particular type of problems at hand. The proposed methodology is generic and can be applied to all types of optimization problems as long as they involve large‐scale finite element simulations. The numerical tests of the present study are performed on sizing optimization of skeletal structures. The numerical tests demonstrate the computational advantages of the proposed parallel strategies, which become more pronounced in large‐scale optimization problems. Copyright © 2003 John Wiley & Sons, Ltd.