AUTOMATIC DESIGN OF ARTIFICIAL NEURAL NETWORKS

AUTOMATIC DESIGN OF ARTIFICIAL NEURAL NETWORKS by Erick Jaimes
English | 2022 | ISBN: N/A | ASIN: B0B9T66WRB | 255 pages | MOBI | 0.40 Mb
The main objective of this work is to meet the needs of a short-term gold market operator, which refers to predicting in the best possible way the value of gold in the following days. In this case, a forecast horizon of 22 business days was chosen (approximately one month of negotiation), for which an ANN was obtained that required only the data series of the preceding five months . The results of the training with the error measures of the trained RNA architectures are shown in Table 1 , where R designates the amount of time delays considered and Hi the number of neurons in the hidden layer. The series of data used for this process corresponds to the closing prices between June 30, 2014 and December 30, 2014. The first columns of Table 1 show that the RNA can model the price of gold in a satisfactory way. using its price series as the only input variable. The best result was obtained in the 1/5/1 configuration corresponding to five neurons in the hidden layer and a time delay in the input layer , in which there are low errors in training and prediction (outside the sample).