Unveiling the impact of causal variables in industrial demand forecasting: A comparative analysis between traditional and machine learning models

Make decisions in supply chain management requires several different information and its accuracy plays a key role in the relative process.Forecasting demand is a major challenge in this scenario as well as improving and developing forecasting models.In the same direction, this article aimed to model, forecast, compare and define among many different models, if causal variables really improve performance metrics as MSE, RMSE and MAPE to define which model is most suitable for mid-term forecasting.Firstly, an extensively literature review was performed in order to provide theoretical background about time series, parametric and nonparametric models and performance measurements.Secondly, several models were fit to the available data and future steps forecasted.At least, performance metrics were evaluated and a better model choose to forecast data.Causal variables performed an important improvement in machine learning models, but not difference in traditional models as SARIMA.In general, comparing error metrics for 6 months forecasting, a SARIMA model overperformed all other tested, even hybrid solutions recently discussed in literature.As a contribution the article provides wide discussion about parametric and nonparametric AI based models as well as a framework to help both researchers and practitioners to choose models in similar supply chain management conditions.