Demand Forecasting for Retail Using Three-S Temporal Fusion (3STF) Network
Open Access DepositedA Selective State Aware Time Series Fusion Memory Model
Retail is a low-margin and high-velocity business. That faces frequent product turnover and must continuously plan for on-shelf product availability. Retailers use demand forecasting to manage these challenges and invest significant resources to get it right. However, there is a high level of error rates in product demand forecasting, which leads to cost overruns, lost sales, extra inventory, and dissatisfied customers .Considering the challenge that error rate is quite high with product demand forecasting that results in higher costs for retailers, lost sales when customers can't find the product they need, extra inventory with anticipation of high demand, and overall customer dissatisfaction. These challenges make demand forecasting very important for retailers. The root case of these issues can be attributed to products that show overlapping trends and seasonality, impact of seasonal promotions, price changes, and calendar and weather-related features. This praxis introduces the Three-S Temporal Fusion (3STF) network. This new model makes use of learnable temporal embeddings, a dual-head forecasting module, and a selective-state scan function to capture linear and nonlinear patterns in the data. Combined, these modules help make the model generalize across a variety of demand patterns. The performance and outcome of this praxis provide ample proof that this new model can improve the demand forecasting accuracy for retailers.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.