Improving Sales Forecasting Accuracy: A Tensor Factorization Approach with Demand Awareness

Because of the accessibility big data collections from consumers, products, and stores, advanced sales forecasting capabilities have drawn great attention many businesses, especially those in retail, because importance decision making. Improvement accuracy, even by a small percentage, may substantial impact on companies’ production financial planning, marketing strategies, inventory controls, supply chain management. Specifically, our research goal is to forecast each product store near future. Motivated tensor factorization methodologies for context-aware recommender systems, we propose novel approach called temporal latent factor forecasting, or ATLAS short, which achieves accurate individualized predictions building single model across multiple stores products. Our contribution combination framework (to leverage information products), new regularization function incorporate demand dynamics), extrapolation into future time periods using state-of-the-art statistical (seasonal autoregressive integrated moving-average models) machine-learning (recurrent neural networks) models. The advantages are demonstrated eight category sets collected Information Resources, Inc., where analyze total 165 million weekly transactions over 15,560 products more than 1,500 grocery stores. Summary Contribution: Sales has been task long-standing importance. Accurate provides critical managerial implications making operations. accuracy management, among other things. This paper proposes computational (machine-learning-based) thus positioned directly at intersection computing business/operations research.