SUPPLY CHAIN AND DEMAND DYNAMICS: CAN AI BE A DRIVING FORCE?

Abstract

With today’s global economy, the complexity of supply chains and the potential for their disruption are escalating, which highlights the need for companies to be agile and data-driven in making decisions. This research investigates how to apply Deep Learning (DL) methods to increase supply chain resilience, improve demand forecasting, and achieve operational efficiency. A quantitative approach was adopted to analyze over 1500 records of the supply chain data in Saudi Arabia, which include historical demand, inventory-level, shipment characteristics, and risk metrics. To preserve the most important information, we performed data pretreatment that included normalization, trend- and domain-informed imputation of missing values, and Principal Component Analysis (PCA) to reduce dimensions. The proposed Adaptive Electromagnetic Field Optimized Attention-enriched Memory Networks (AEFO-Att-MN) made use of the treated data for input. The networks used (LSTM) networks to take into account long term temporal dependencies along with attention mechanisms to highlight significant features and (AEFO) to optimize feature weights and model parameter. The framework was made with Python, TensorFlow, and PyTorch. The performance test showed that the model's ability to make accurate predictions has improved a lot. The Root Mean Squared Error (RMSE) is 0.412, the Mean Absolute Error (MAE) is 0.365, and the R² is 0.862. The results showed that there were fewer times when there was too much stock or not enough stock, better management of lead times, and shipments that were at risk are found before they happen. Moreover, the results showed that using advanced preprocessing and AEFO-Att-MN techniques together made forecasts much more accurate and made it easier to deal with unexpected events. This is useful information for AI-driven supply chain management.