A neural network model for crude oil price forecasting

Iran is one of the top five important countries in the world that have rich oil reserves. Exchange incomes produced by oil exports play an important role in country’s budget. Therefore, the studies and researches in fields that are related to oil economics have great privilage. Today, there is a plentiful interest in use of artificial intelligence methods especially neural networks for improving financial decision-making procedures. The main advantage of neural networks is their ability for learning from past experiments. They improve their efficiency by network training. So in this research, a neural network model is developed for forecasting monthly crude oil price using supervised learning. In development process of this model, effect of network structure types, technical inputs types, fundamental inputs, number of input layer neurons, number of hidden layers and neurons, transfer function of layers, data preprocess, divide of data to train and test sets and improved learning algorithm types are investigated. A three layers feed forward network with N9-2-8-1 structure is selected as the best price-forecasting model. This network has mean absolute error equal to 0.74$ on train set and 0.71$ on test set.