Catálogo de publicaciones - libros

Financial ratios are commonly employed to measure a corporate financial performance. In recent years a considerable amount of research has been directed towards the analysis of the predictive power of financial ratios as influential factors of corporate stock market behavior. In this paper we propose a constraint-based evolutionary classification tree (CECT) approach that combines both the constraint-based reasoning and evolutionary techniques to generate useful patterns from data in a more effective way. The proposed approach is experimented, tested and compared with a regular genetic algorithm (GA) to predict corporate financial performance using data from Taiwan Economy Journal (TEJ). Better prediction effectiveness of CECT approach is obtained than those of regular GA and C5.0.

A genetic algorithm with chromosome-repairing scheme (CRS) is proposed in this paper to solve the polygonal approximation problem. Different from the existing approaches based on genetic algorithms, the proposed algorithm adopts variable-length chromosome encoding for reducing the memory storage and computational time, and develops a special crossover named gene-removing crossover for removing the redundant genes. It is known that Genetic operators may yield infeasible solutions, and it is generally difficult to cope with them. Instead of using the penalty function approach, we propose a chromosome-repairing scheme to iteratively add the valuable candidate gene to the chromosome to deal with the infeasible solution and an evaluating scheme for the candidate genes. The experimental results show that the proposed CRS outperforms the existing approaches based on genetic-algorithms, ant-colony-optimization and tabu-search.

In large scale industry systems, especially in chemical process industry, large amounts of variables are monitored. When all variables are collected for fault diagnosis, it results in poor fault classification because there are too many irrelevant variables, which also increase the dimensions of data. A novel optimization algorithm, based on a modified binary Particle Swarm Optimization with mutation (MBPSOM) combined with Support Vector Machine (SVM), is proposed to select the fault feature variables for fault diagnosis. The simulations on Tennessee Eastman process (TEP) show the BMPSOM can effectively escape from local optima to find the global optimal value comparing with initial modified binary PSO (MBPSO). And based on fault feature selection, more satisfied performances of fault diagnosis are achieved.

The problem of automatic classification of ultrasound images is addressed. For texture analysis of ultrasound images quantifiable indexes, called features, are used. Classification was performed using Gaussian mixture model based on Bayes classifier. The common problem of texture analysis is a feature selection for classification tasks. In this work we use genetic algorithms for a feature subset selection. Total number of 387 features was used, consisting of spatial an co-occurance statistical texture features (proposed by Muzzolini and Haralick). The classification infers between healthy thyroid gland and thyroid gland with chronic inflammation.

The performance of the Nearest Neighbor classifier drops significantly with the increase of the overlapping of the distribution of different classes. To overcome this drawback, we propose to simulate the physical process of gravitational collapse to trim the boundaries of the distribution of each class to reduce overlapping. The proposed simulated gravitational collapse(SGC) algorithm is tested on 7 real-world data sets. Experimental results show that the nearest prototype classifier based on SGC outperforms conventional NN and k-NN classifiers.

This paper presents the rule generation method based on evolutionary computation and rough set, which integrates the procedure of discretization and reduction using information entropy-based uncertainty measures and evolutionary computation. Based on the definitions of certain rules and approximate certain rules, the paper focuses on the reduction by meanings of evolutionary computation. Experimental results reveal that the proposed method leads to better classification quality and smaller number of decision rules comparing with other methods.

In this contribution, we propose and analyze three evaluation functions (contrast functions in Independent Component Analysis terminology) for the use in a genetic algorithm (PNL-GABSS, Post-NonLinear Genetic Algorithm for Blind Source Separation) which solves source separation in nonlinear mixtures, assuming the post-nonlinear mixture model. Blind source separation refers to the problem of recovering a set of unknown sources from another set of mixtures directly observable and little more information about the way they were mixed. Assuming statistical independence as the assumption to obtain the original sources we can apply ICA (Independent Component Analysis) as the technique to recover the signals. In order to analyze in practice the performance of the chosen fitness functions in our proposed algorithm, we applied ANOVA (Analysis of Variance) to the results, showing the validity of the three approaches.

In this paper, we present a novel soft decision mixture model for image segmentation. This model adopts the soft decision classify into gaussian mixture model to represent the probability distribution of the observed image feature. The model for the underlying true context images is designed to serve as prior contextual constraints on unobserved pixel labels in term of markov random field model. Experiments with synthetic image and real image show that the use of soft decision mixture model definitely improves the quality of the segmentation results for noisy images and results in reduced classification errors in the interior area of the region.

The accuracy of a real-time digital filter frequency response is affected by the finite word length (FWL) constraint of coefficients used in its implementation. In this paper, we consider the FWL problem in regard to the finite impulse response (FIR) digital filters. Some theoretical issues and statistical error bound conditions of the maximum deviation between the exact and the approximate magnitude responses are also considered. We use real-valued genetic algorithms (GA) as an optimisation tool and derive results for the maximum error bounds and error deviation due to FWL effects for a number of design examples. Finally, a comparison is drawn between the simply rounded, the GA optimised, integer programming and the simple hill climber methods.

DSPs provide high performance and low cost through their use of specialized hardware features. One feature commonly found in DSPs is the dual data memory banks to offer high memory bandwidth. However, it poses problems for C compilers, which are mostly not capable of assigning variables between banks. In this paper, an immune genetic algorithm for variable partition between data banks is presented to maximize the benefit of this feature. In our approach, the reduced interference graph of variable accesses is constructed, the potential variable partitions are represented as antibodies and the vaccines are abstracted; then through some operations including adaptive vaccination, immune selection and so on, the antibodies can converge at optimal variable partitions. Experimental results show that our algorithm is superior to previous works in terms of performance and code size.