This study presents a novel approach to the demand forecasting problem in the steel industry, utilizing a multivariate time series forecasting model, specifically the Long Short-Term Memory (LSTM) neural network. The steel industry places a high value on demand forecasting due to the complexity of its production processes, increasing lead times, and market uncertainty. The proposed methodology uses machine learning techniques to develop a multivariate LSTM model and evaluates its performance using various indicators. Additionally, to align with the unique characteristics of the steel industry, we performed feature selection based on the insights of industry experts. In conclusion, the LSTM model developed in this study outperforms existing demand forecasting methods. Specifically, it accurately reflects complex market trends and the characteristics of the steel industry, enabling more accurate demand forecasting. This study presents a novel solution to the demand forecasting problem in the steel industry, which is expected to enhance the efficiency of production planning and inventory management.