Application of genetics algorithms for estimating the parameters of a Cox regression model
Douglas Rivas, José Luciano Maldonado, Rafael Borges, Gerardo Colmenares · 2006
This paper, involved in the area of evolutive computing, presents the development of a Genetic Algorithm in finding the optimal parameters of a Cox Regression Model for patients of the Service of Peritoneal Dialysis of the “Hospital Clinico Universitario de Caracas”between 1980 and 2002, performed by Borges (2002, 2005). The technique of the genetic algorithms is used as a method for finding a better estimation of the parameters of the Cox Model than the obtained by the classical optimization methods. The algorithm was completed programmed in the language C++, using a modular programming design, considering every element of the genetic algorithms. The main characteristics of the algorithm are: a) the initial population, of 10 subjects, is generated randomly between a range of values, this range was obtained after several essays, b) the adjustment function was based in the Akaike Information Criteria (AIC), c) the selection of the subjects to be reproduced was done by tournament, d) the multipoint operator for the crossing and, the mutation was done to all the genes of one part of the chromosomes of the population. The developed algorithm was useful to obtain the estimation of the Cox Regression Model and with better AIC values than the obtained by the classical methods.