FOREX Prediction Using An Artificial Intelligence System

Abstract

The purpose of this study is to examine the use and applicability of an artificial intelligence system in predicting changes in foreign currency exchange rates. There are algorithms available for that purpose and this study compares several of these algorithms for efficiency and accuracy. This comparison was carried out through the use of the Metlab computer software program. The multi-layer back-propagation neural network was chosen for this research. We use feed-forward topologies, supervised learning and back-propagation learning algorithms on the network. This program allows for training neural networks, thereby producing predictions of future foreign currency exchange rates. This paper builds a model for pattern recognition of foreign currency exchange rate trends. The methodology used in this paper was successful in that neural networks were successfully trained and predictions of future foreign currency exchange rates were produced. A total of eleven algorithms and different exchange rates were compared and tested through the neural network training procedure. The Levenberg-Marquardt algorithm is best suited to deal with a function approximation problem where the network has up to several hundred weights, and the approximation must be very accurate. Over all, of the algorithms considered, the Levenberg-Marquardt algorithm appears to be the most appropriate for the purposes of this paper.