Abstract

In the industrial context, steel is a broadly-used raw material with applications in many different fields. Due to its high impact in the activity of many industries all over the world, forecasting its price is of utmost importance for a huge amount of companies. In this work, non-linear neural models are applied for the first time to different datasets in order to validate their suitability when predicting the price of this commodity. In particular, the NAR, NIO and NARX neural network models are innovatively applied for the first time to forecast the price of hot rolled steel in Spain. Besides these variety of models, different datasets consisting of a set of heterogenous variables from the last seven years and related to the price of this commodity are benchmarked and analyzed. The results showed that NARX is the best performing model when the price of raw materials used to produce steel and the stock market prices of three major global steel producing companies are employed as input to this predictive model. Consequently, this result may boost the application of Machine Learning in companies, in order to schedule the supplying operations according to the price forecasting.