A Manager-Assisted Hybrid Decision Support System Using Advanced AI Models for Demand Forecasting and Inventory Risk Management in Fashion Manufacturing and Distribution

Source abstract

Fashion enterprises face demand uncertainty from seasonality, short product life cycles, and dense color-size variation: excessive ordering enlarges aged-inventory risk, whereas insufficient ordering causes stockouts. This study develops and validates a manager-assisted hybrid decision support system (DSS) integrating demand forecasting, reorder-candidate identification, aged-inventory risk detection, and ranking-based prioritization. Operational data from a Korean fashion enterprise (4,130,603 sales and 5,065,027 order records) were reconstructed into a style-level monthly panel of 263,260 observations and split chronologically into training (2012–2014), validation (2015), and test (2016) periods to prevent leakage. Three baselines and nine data-driven models, Random Forest, Histogram Gradient Boosting, LightGBM, a two-stage Hurdle LightGBM, LSTM, Transformer, TimeMixer, TimeXer, and CrossLinear, were benchmarked, their outputs then combined with current-month sales signals and an inventory-comparison rule inside the DSS. On the 2016 test data, the DSS attained a Weighted Absolute Percentage Error of 0.6477, reorder F1-score of 0.7983, Hit@50 of 47.04%, and NDCG@50 of 75.12%, a balance no single model achieved. A policy simulation indicated that DSS-guided ordering could cut inbound order quantity by 25.9% and long-term inventory risk by 23.0%, at the cost of a 39.5% rise in shortage quantity. Forecasting accuracy alone is therefore insufficient for deploying artificial intelligence in inventory decisions.