Steel Price Forecasting Using Long Short-Term Memory Network Model

Abstract

In this study, a steel price forcasting model has been developed by using the Long Short-Term Memory Network Model (LSTM) which is a customized model of recurrent neural network (RNN) architecture. The 10-years stock closing prices of the 50 largest iron and steel companies traded in the world stock exchanges and 10-years data of the scrap metal price obtained from the London metal exchange (LME) on the same day and dates combined as time series for using model training and testing stages. As a result of the forcasting made with the UKDHA model, which is designed to have 1 lstm, 7 dense layers, the best forcasting result was obtained from the forward 5-day frocasting model with the correlation coefficient R =3D 0.8559, MSE value 0.0026 and MAE value 0.0383. © 2019 IEEE.