Fuzzy Cognitive Maps With Genetic Algorithm for Goal-Oriented Decision Support

Abstract

Fuzzy cognitive maps are signed directed graphs used to model the evolution of scenarios with time. FCMs can be useful in decision support for predicting future states given an initial state. Genetic algorithms (GA) are well-established tools for optimization. This paper concerns the use of FCMs in goal-directed analysis of scenarios for aiding decision making. A methodology for GA-based goal-directed analysis is presented. The search for the initial stimulus state, that over time leads to a target state of interest, is optimized using GA. This initial state found can be used to answer the question – what course of events leads to a certain state in a given scenario?

Improving fuzzy cognitive maps learning through memetic particle swarm optimization

“…A learning algorithm modifies further the weights of the FCM, such that the achieved stable state is desirable, with the concepts and weights lieing within the bounds posed by the experts. Established learning schemes are based either on the Hebbian rule for unsupervised neural networks training (Kosko 1997; or on evolutionary computation and swarm intelligence schemes (Khan et al 2004;Koulouriotis et al 2003;Papageorgiou et al 2005;Parsopoulos et al 2004b;Stach et al 2005). In the latter case, properly defined objective functions that penalize weight settings corresponding to undesirable steady states, are used Parsopoulos et al 2004a,b).…”

“…Learning methods that utilize PSO and DE have been introduced in Papageorgiou et al (2005), Parsopoulos et al (2004b) and Petalas et al (2005Petalas et al ( , 2007a. GAs (Khan et al 2004;Stach et al 2005) and Evolution Strategies (ES) (Koulouriotis et al 2001) have also been successfully used.…”

“…These objectives are defined by the related AAPM and ICRP protocols (Khan 1994;Wells and Niederer 1998;Willoughby et al 1996) for the determination of accepted dose levels for each organ and part of the human body. The ranges of the weights for the supervisor-FCM model are given in Table 2.…”

“…There are just a few well-studied learning schemes in the literature and they can be classified in two categories. The first consists of approaches based on unsupervised neural networks training methods (Kosko 1997;, while the latter refers to evolutionary computation and swarm intelligence approaches (Khan et al 2004;Koulouriotis et al 2003;Papageorgiou et al 2005;Parsopoulos et al 2004b;Stach et al 2005).…”

“…Also, an MPSO approach has been applied for FCMs learning (Petalas et al 2005(Petalas et al , 2007a, with preliminary results indicating its viability. Different learning methods, such as Differential Evolution (DE), Genetic Algorithms (GAs), and Evolution Strategies, have been also proposed and applied successfully for FCMs learning (Khan et al 2004;Koulouriotis et al 2001;Papageorgiou et al 2005;Parsopoulos et al 2004b;Petalas et al 2005Petalas et al , 2007aStach et al 2005).…”