Computational intelligence techniques for modelling an economic system

Mastery of the practice of economic modeling has long attracted the interests of economists, government bureaucrats, political theoreticians and scientists alike. In today's global socio-political environment, economics has become an important and central feature of the determinants that shape the policies, outlook and character of modern nation states. Economics extends far beyond its traditional formulation and determines even international relations and politics. All of these underpin the importance of tools that can be used for modeling an economic system. This makes the development of an accurate model of any nation's economy an interesting research and engineering problem. The authors of this paper present an ensemble of the results of two computational intelligence techniques in an attempt to solve this engineering problem. The techniques used are the Multi-layer perceptron (MLP) model and Rough set theory. Outputs of each method are combined to give a singular output. Rough set theory has, as its base, imperfect data analysis and approximation. The theory is used to extract a set of reducts and a set of rules based on relationships deduced from 10 attributes that influence the direction of the percentage change in the gross domestic product of the South African economy. The data used spans from 1980 to the year 2010. The MLP model developed consists of a single hidden layer and several hidden units. The optimal selection of the number of hidden layers, number of hidden units and values of weights is determined by the particle swarm optimization algorithm. The model gave a prediction accuracy of 86.8 %.