AI-Driven Raw Material Demand Forecasting: Towards Project Management Practices

Abstract

Accurate forecasting of raw material demand is critical for effective decision-making in project management, particularly in the construction sector. Demand fluctuations in this industry can severely disrupt workflows and cause project delays. Reliable demand predictions help maintain operational stability and reduce risks, especially under uncertain conditions. In such environments, maintaining appropriate Target Stock Levels (TSL) and setting effective Reorder Points (RP) are essential to ensure project continuity and customer satisfaction. This study investigates the potential of Artificial Intelligence (AI) to enhance demand forecasting accuracy within service-based supply chains. Five forecasting models were evaluated: four machine learning approaches—Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), Random Forest (RF), and Gaussian Regression (GR)—and one traditional statistical method, the Autoregressive Integrated Moving Average (ARIMA) model. The models were trained and tested using historical raw material consumption data collected from a construction project over a four-year period (2019–2023). The results show that the Machine Learning Models significantly outperformed the ARIMA model in terms of predictive accuracy. The coefficients of determination (R²) were 0.93 for LSTM, 0.91 for XGBoost, 0.89 for GR, and 0.88 for RF, compared to 0.75 for ARIMA. Among all models, LSTM achieved the highest forecasting accuracy and the lowest deviation on the test dataset. Its implementation for the 2024 planning horizon led to substantial inventory optimization, reducing overstock volumes by 66.5%. This improvement translated into significant cost savings and enhanced the overall efficiency of material management and decision-making processes.