Hybridizing Genetic Algorithms with ICA in Higher Dimension Juan Manuel G´ orriz 1 , Carlos G. Puntonet 2 , Mois´ esSalmer´on 2 , and Fernando Rojas Ruiz 2 1 E.P.S. Algeciras, Universidad de C´adiz Avda. Ram´on Puyol s/n, 11202 Algeciras C´adiz, Spain juanmanuel.gorriz@uca.de 2 E.S.I., Inform´atica, Universidad de Granada C/ Periodista Daniel Saucedo, 18071 Granada, Spain {carlos,moises}@atc.ugr.es Abstract. In this paper we present a novel method for blindly sep- arating unobservable independent component signals from their linear mixtures, using genetic algorithms (GA) to minimize the nonconvex and nonlinear cost functions. This approach is very useful in many fields such as forecasting indexes in financial stock markets where the search for independent components is the major task to include exogenous in- formation into the learning machine. The GA presented in this work is able to extract independent components with faster rate than the previ- ous independent component analysis algorithms based on Higher Order Statistics (HOS) as input space dimension increases showing significant accuracy and robustness. 1 Introduction The starting point in the Independent Component Analysis (ICA) research can be found in [1] where a principle of redundancy reduction as a coding strategy in neurons was suggested, i.e. each neural unit was supposed to encode statistically independent features over a set of inputs. But it was in the 90´s when Bell and Sejnowski applied this theoretical concept to the blindly separation of the mixed sources (BSS) using a well known stochastic gradient learning rule [2] and originating a productive period of research in this area [3–6]. In this way ICA algorithms have been applied successfully to several fields such as biomedicine, speech, sonar and radar, signal processing, etc. and more recently also to time series forecasting [7], i.e. using stock data [8]. In the latter application the mixing process of multiple sensors is based on linear transformation making the following assumptions: 1. the original (unobservable) sources are statistically independent which are related to social-economic events. 2. the number of sensors (stock series) is equal to that of sources. 3. the Darmois-Skitovick conditions are satisfied [9]. C.G. Puntonet and A. Prieto (Eds.): ICA 2004, LNCS 3195, pp. 414–421, 2004. c  Springer-Verlag Berlin Heidelberg 2004