Gregorio Fidalgo Valverde, Alicja Krzemień, Pedro Riesgo Fernández, Francisco Javier Iglesias Rodríguez, Ana Suárez Sánchez
This paper presents a twelve-month forecast of copper price time series developed by means of Generalized regression neural networks with optimized predictor variables. To achieve this goal, in first place the optimum size of the lagged variable was estimated by trial and error method. Second, the order in the time series of the lagged variables was considered and introduced in the predictor variable. A combination of metrics using the Root mean squared error, the Mean absolute error as well as the Standard deviation of absolute error, were selected as figures of merit. Training results clearly state that both optimizations allow improving the forecasting performance.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados