DEVELOPMENT OF A TOOL FOR DEMAND FORECASTING IN A PREMIUM TUBULAR SOLUTIONS COMPANY USING MACHINE LEARNING

Demand forecasting is central to industrial planning in segments of high variability, such as premium tubular solutions. A tool based on the XGBoost algorithm is proposed for monthly forecasting of the demand of heavy vehicles associated with pipe consumption in a Brazilian steel company. The methodology includes training (2017–2023) / test (2024+) division, lag variables (1, 2, 3, 6 and 12 months), treatment of the pandemic period as structural shock via indicator variable, regularization and quarterly evaluation. The results show that the model captures the post-pandemic structural downward trend with reasonable accuracy, with MAE of 1,277 tons/quarter, RMSE of 1,506 tons, WAPE of 18.63%, and MAPE of 21.91%. It is concluded that the integration between business knowledge, treatment of structural changes and careful choice of metrics is decisive for the construction of robust models in industrial planning.