engleski

Neural networks for time-series predictions in finance and investing

According to research of many authors, almost all problems can be solved more efficient by neural networks than by traditional modelling and statistical methods. Artificial intelligence with neural networks gives the possibility to improve classical methods with its capability of learning, higher degree of robustness and fault tolerance. The paper is concerned on usage of neural networks in domain of finance and investing. Various authors are compared and their results in neural network application in area of finance and investing are presented. In our research, we tried to test and evaluate several different architectures of backpropagation neural network algorithm on profit prediction problem. Given results show that the architecture of neural networks that best predicts the future values of profits with minimum root mean square error is three layer network with 9 neurons in input and hidden layer and one neuron in output layer, with learning parameter of 0.2. Future research can concentrate on other evaluating measures and types of networks.

neural networks; backpropagation algorithm; time series prediction; profit

nije evidentirano