Machine Learning for Sales Forecasting in Industrial Environments: A Systematic Literature Review of Input Factors and Applied Methods

Accurate sales forecasting is crucial for manufacturing companies, as it significantly impacts inventory, capacity utilization, lead times, and delivery performance. However, these forecasts are subject to considerable uncertainty driven by economic, political, and environmental dynamics. Under such conditions, the reliability of established forecasting approaches is reduced. Concurrently, digitalized supply chains yield large data sets that can enhance the prediction quality of sales forecasts. Machine learning (ML) facilitates the extraction of underlying patterns from these diverse data sets, enabling data-driven and more accurate predictions. The existing scientific literature introduces numerous ML-based approaches for sales forecasting. This study presents the findings of a systematic literature review on the characteristics of ML-based sales forecasts, utilized input factors, applied methods for both feature selection and regression analysis, as well as interpretability approaches from the field of explainable artificial intelligence. Thus, this study highlights existing research gaps, particularly concerning the selection and quantification of relevant input factors for ML-based sales forecasting in industrial environments.