Applying intelligent methods for analysis and decision making is becoming popular. Modern computers with powerful processors have enabled machine learning algorithms to be used for system identification or at least their governing concepts. One common problem in decision making is the prediction of future behavior of a system. To have a good and accurate prediction, first we need to be assured of the predictability of the system. In this thesis after being assured of the predictability of system using intelligent tests, the possibility of random behavior is rejected. Used tests include independence test, probability of return to zero, first return to zero and Brock-Dechert-Scheinkman algorithm. In these tests we compare system outputs vs. random walks' outputs. Then discussion is continued about volatility and chaos to find a reason for rejection of random outputs and predictability. Using Singular Spectrum Analysis and predicting each component using neural networks, the prediction is performed. Locally linear fuzzy neural networks show the best result. To make a better prediction we should select the range of recurrence, False Nearest Neighbors and Neighbors Cardinality methods are used to select these limits. Some foreign exchange rates as important macroeconomic indices are chosen as the test bench for algorithms. The results of this research show non-random and deterministic data and also lower sensitivity to initial conditions in these macroeconomic time s eries .
