ML-DRIVEN DEMAND FORECASTING IN LEGACY ERP ENVIRONMENTS

Abstract

Accurate demand forecasting is essential for efficient planning, inventory management, and cost reduction in manufacturing and supply chain activities. But a lot of businesses still use old Enterprise Resource Planning (ERP) systems that were made for processing transactions, not for advanced analytics. Legacy ERP systems, like QAD, apply forecasting methods that tend to struggle with newer changes in supply chains. This research investigates whether demand forecasts can be improved using ML (Machine Learning) models such as XGBoost (eXtreme Gradient Boosting) and LSTM (Long Short-Term Memory), which have been tested on the M5 Forecasting dataset. Data preparation and feature development allowed ML models to outsmart moving averages and ARIMA (AutoRegressive Integrated Moving Average) in following the details of the demand response. The findings demonstrate that with ML, accuracy in forecasting increases in the QAD ERP system, helping with strategizing and planning inventory choices without upgrading the leading ERP. This shows how to improve legacy ERP systems by using data insights at any scale.