Optimization of steel demand forecasting with complex and uncertain economic inputs by an integrated neural network–fuzzy mathematical programming approach

This study presents an integrated approach to optimize steel consumption forecasting. The unique nature of this approach lies in the integration of the artificial neural network (ANN), fuzzy linear regression (FLR), and conventional linear regression (CLR) approaches. Hence, it can be easily applied to complex, ambiguous, and certain environments due to its flexibility. This approach is applied to forecast steel consumption in USA (the top 10 of the word in steel per capita consumption) and Iran (the top 10 of the Middle East in steel per capita consumption). In order to capture the potential complexity, uncertainty, and linearity relation between explanatory variables and steel consumption function, ANN, FLR (including eight distinct FLR approaches), and CLR approaches are applied to both cases. Next, the analysis of variance (ANOVA) is performed on the forecasts of the selected FLR, selected ANN, and actual values for the cases of USA and Iran, separately. The achieved results from ANOVA and MAPE results for the case of USA demonstrated the considerable superiority of selected ANN over selected FLR and CLR regarding to the nonlinear and complex nature of steel consumption function in USA. However, the superiority of selected FLR over its rivals for steel consumption forecasting in Iran is referred to unstable economic status of Iran. Finally, in order to verify and validate the results, a sensitivity analysis through changing train and test data is carried out. This is the first study that presents an ANN–FLR–CLR approach for steel consumption forecast capable of handling complexity, nonlinearity, and uncertainty.