Sheida Torbat - Department of Industrial and Systems Engineering, Isfahan University of Technology (IUT), Isfahan, Iran
M. Khashei - Department of Industrial and Systems Engineering, Isfahan University of Technology (IUT), Isfahan, Iran
M. Bijari - Department of Industrial and Systems Engineering, Isfahan University of Technology (IUT), Isfahan, Iran

Steel consumption is a critical factor affecting pricing decisions and a key element to achieve sustainable industrial development. Forecasting future trends of steel consumption based on analysis of nonlinear patterns using artificial intelligence (AI) techniques is the main purpose of this paper. Because there are several features affecting target variable which make the analysis of relationships between variables difficult, so it is needed  forintelligence tools in order to recognize the nonlinear patterns and select the most effective features as final input variables. Generally, in the real world problems, a nonlinear pattern recognition and feature clustering usually involve high-dimensional features that make the clustering problem complex. In fact without feature subset selection and dimensionality reduction, both training accuracy and generalization capability of forecasting models will be significantly reduced. In this paper, artificial neural networks are applied in order to profit from unique advantages of both forecasting and clustering power of the artificial neural networks to create an efficient and accurate model in high-dimensional situations. This way, at first, self-organizing map (SOM) is applied as an unsupervised clustering technique to detect nonlinear patterns between explanatory variables and to determine final effective input variables. Then the best multilayer perceptron is designed by these variables in order to forecast future trends of steel consumption. The empirical results of Iran's steel consumption forecasting confirm that the proposed model exhibits effectively improved forecasting accuracy in comparison to traditional feature selection methods such as forward and backward.

Self-organizing maps (SOMs), Nonlinear Pattern recognition, Feature clustering, Dimensionality reduction, Artificial neural networks (ANNs), Steel consumption forecasting