Choosing Variables with a Genetic Algorithm for Econometric models based on Neural Networks learning and adaptation

The mixture of two already known soft computing techniques, like Genetic Algorithms and Neural Networks (NN) in Financial modeling, takes a new approach in the search for the best variables involving an Econometric model using a Neural Network. This new approach helps to recognize the importance of an economic variable among different variables regarding econometric modeling. A Genetic algorithm constructs a set of working neural networks, evolving the inputs given to each NN as well as its internal architecture. An input subset is chosen by the genetic algorithm from a multiple variable set, due to the NN training results from this given input. At the end of the evolutionary process, the best given inputs for a specific neural network architecture are obtained. The experimental results revealed an improvement of 80% in the NN learning performance of the Econometric model. At the same time it reduces the model complexity by 46%, without large computer resources being used during the evolutionary process.