This part has two main sections.
The first section is a theoretical part, dealing with the theoretical background of the main application field of this thesis, application field which is time series prediction. This section starts with a definition of the time-series prediction problem, then goes through a brief history of time-series prediction research and then presents some of the most popular tools in this particular field, one of the most successfull ones being Neural Networks.
The second section deals with the computer simulations, the experiments and the results and discussion for this problem. This section starts with the definition of the particular problem treated here, stock exchange forecasting, and a short explanation of the stock exchange mechanism. After these frame setting explanations, the neural network is built from scratch under the very eyes of the user, starting with the data processing and selection, the building of a global neural network, taken into consideration other external factors implied in the particular field of study. After the processing cleaning steps, the remaining time series can serve as direct input to a ANN. I describe here a mathematically optimized algorithm used for training, the network suitable for this algorithm, the data used for simulation, and then the simulation results and discussions.