Abstract
Accurate forecasting of monthly steel production is essential for maintaining efficient industrial operations and ensuring timely delivery of coil and rebar products to customers. This paper presents a comparative evaluation of two data-driven prediction methods, the Adaptive Neuro-Fuzzy Inference System (ANFIS) and the Long Short-Term Memory (LSTM) network. The dataset is based on 48 months of historical production data. To ensure a fair comparison, the input data were normalized, and the parameters of each model were systematically optimized. The model performance metrics included the Root Mean Square Error (RMSE), Mean Relative Error (MRE), and the coefficient of determination (R²). The results showed that the ANFIS model achieved superior forecasting performance, with RMSE = 236.93, MRE = 0.0081, and R² = 0.9947. However, the LSTM model produced relatively higher errors, with RMSE = 696.13, MRE = 0.0234, and a lower R² = 0.9819. Moreover, a one-way Analysis of Variance (ANOVA) statistical analysis confirmed that the difference between the two models was statistically significant (F = 45.59, p = 1.4997×10⁻⁹), indicating that the superior performance of ANFIS is statistically reliable. The results suggest that ANFIS is more effective for the limited available production dataset, while LSTM performance is expected to improve with larger sequential datasets.