Jose Crispin Hernandez

A memetic algorithm for gene selection and molecular classification of cancer

Choosing a small subset of genes that enables a good classification of diseases on the basis of microarray data is a difficult optimization problem. This paper presents a memetic algorithm, called MAGS, to deal with gene selection for supervised classification of microarray data. MAGS is based on an embedded approach for attribute selection where a classifier tightly interacts with the selection process. The strength of MAGS relies on the synergy created by combining a problem specific crossover operator and a dedicated local search procedure, both being guided by relevant information from a SVM classifier. Computational experiments on 8 well-known microarray datasets show that our memetic algorithm is very competitive compared with some recently published studies.

SVM-Based Local Search for Gene Selection and Classification of Microarray Data

This paper presents a SVM-based local search (SVM-LS) approach to the problem of gene selection and classification of microarray data. The proposed approach is highlighted by the use of a SVM classifier both as an essential part of the evaluation function and as a “provider” of useful information for designing effective LS algorithms. The SVM-LS approach is assessed on a set of three well-known data sets and compared with some best algorithms from the literature.

A study of crossover operators for gene selection of microarray data

Classification of microarray data requires the selection ofa subset of relevant genes in order to achieve good classification performance.Several genetic algorithms have been devised to perform thissearch task. In this paper, we carry out a study on the role of crossover operatorand in particular investigate the usefulness of a highly specializedcrossover operator called GeSeX (GEne SElection crossover) that takesinto account gene ranking information provided by a Support Vector Machineclassifier. We present experimental evidences about its performancecompared with two other conventional crossover operators. Comparisonsare also carried out with several recently reported genetic algorithms onfour well-known benchmark data sets.

Digital Controller for an Electric Wheelchair Based on a Low-Cost Hardware

This paper presents a digital motion controller of an electric wheelchair by using a low-cost real-time microcontroller unit (MCU) and an own developed power electronic drive. The speed control of the wheels is implemented through a digital pulse-width modulation technique which is varying with the reference generated by a low-cost joystick. Voltage regulators and power converters are implemented to supply correct voltage levels to control cards and to the DC motors from batteries. Experimental results which have been obtained by using the MCU, the mechanical structure of a used electric wheelchair and the developed electronic drive demonstrate that the proposed digital controller meets the user requirements in terms of reliability, good driving and fair economic cost.

Maximum Expectation algorithm and neuronal network base radial applied to the estimate of an environmental variable, evapotranspiration in a greenhouse

This article analyzes the data mining techniques to get the evapotranspiration variable (ETo) in order to control the irrigation system in green houses to optimize resources such as water and fertilizers. The methods used are the Maximum Expectation algorithm (EM) and the neuronal network base radial; such methods estimate the environmental variable from historical data such as: the temperature and moisture values collected from the sensors that are within the green house. These methods predict values from statistical distribution that give us the optimum value for temperature and moisture values occurring at that moment. This data will be compared to determined values for the formulas of the Penman-Monteith model, which has been until now the model with more reliable results.

Recursive feature elimination based on linear discriminant analysis for molecular selection and classification of diseases

We propose an effective Recursive Feature Elimination based on Linear Discriminant Analysis (RFELDA) method for gene selection and classification of diseases obtained from DNA microarray technology. LDA is proposed not only as an LDA classifier, but also as an LDAs discriminant coefficients to obtain ranks for each gene. The performance of the proposed algorithm was tested against four well-known datasets from the literature and compared with recent state of the art algorithms. The experiment results on these datasets show that RFELDA outperforms similar methods reported in the literature, and obtains high classification accuracies with a relatively small number of genes.