A Data-Driven Framework for Demand Forecasting and Inventory Optimization in ERP Using Machine Learning and Ddmrp on FPGA
2026 International Biennial Conference on Advances in Artificial Intelligence and Data Science, IBAAIDS 2026, Bushehr, İran, 27 - 28 Ocak 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/ibaaids66771.2026.11567409
- Basıldığı Şehir: Bushehr
- Basıldığı Ülke: İran
- Anahtar Kelimeler: DDMRP, Demand Forecasting, ERP Systems, Inventory Management, Machine Learning, Multi-Objective Optimization
- Atatürk Üniversitesi Adresli: Hayır
Özet
Combining demand forecasting with inventory planning is still a major challenge for enterprise resource planning (ERP) systems, and the outcome often leads to excessive stock or shortages. We propose a data-driven framework that uses machine learning to predict demand and DDMRP-inspired supply logic to optimize inventory policies. Several forecasting models are evaluated using Walmart weekly sales data from 45 stores, including a simple seasonal baseline model, LightGBM, XGBoost, and CatBoost. The findings show that CatBoost reduces forecasting error by about 31 % compared to the baseline. Weekly forecasts are converted into daily demand scenarios and applied in a stochastic inventory simulation that considers lead times, inventory levels, and shortages. A multi-objective optimization with Pareto front analysis reveals clear trade-offs between service level and inventory investment. Three policies are obtained: Lean, Balanced, and High-Service. The Balanced policy achieves a reasonable compromise between service level and cost, with a 97.9 % supply rate and nearly 50 % fewer shortage days than the Lean policy. Successful deployment of this framework on a Zynq UltraScale+ MPSoC board and its integration with an interactive user interface demonstrate its practical applicability in real-Time industrial environments, supporting stronger ERP decisions.