Predictive analytics for product demand: An end-to-end model assessment

In the fast-evolving domain of supply chain management, accurate demand forecasting is essential to optimize inventory management, reduce costs, and improve customer satisfaction. This research introduces a comprehensive approach to forecasting product demand using a range of machine learning techniques, such as ARIMA, Random Forest, XGBoost, and Simple Exponential Smoothing (SES). Each technique is meticulously evaluated using key performance indicators, including the mean squared error (MSE), the root mean squared error (RMSE), the mean absolute error (MAE) and the R-squared (R2) score. A particular emphasis of this research is on employing a stacking ensemble method to enhance the accuracy of predictions. Additionally, considerable attention is given to the process of feature selection using ANOVA TEST and extraction, which are vital for the models’ ability to effectively interpret and forecast complex sales patterns. The methodologies are tested and validated using a data set that includes sales records from a variety of products and stores, with the objective of accurately forecasting weekly sales volumes.