Thursday 06 March 2025
A new approach to demand forecasting in supply chain management has been proposed, one that leverages the power of graph neural networks (GNNs) to better predict future demands. The traditional methods used in this field rely on simplistic models that fail to capture the complex relationships between nodes in a supply chain network. However, GNNs have shown promise in modeling these relationships and making more accurate predictions.
The researchers behind this study used a dataset called SupplyGraph, which contains information about various products and their supply chains. They divided the data into training, validation, and testing sets and trained three different models: a multi-layer perceptron (MLP), a graph neural network (GNN), and a graph convolutional network (GCN). The MLP is a traditional deep learning model that relies on feedforward connections between nodes, while the GNN and GCN are designed specifically for graph-structured data.
The results of the study show that the GNN outperformed both the MLP and GCN in terms of mean squared error (MSE) and mean absolute error (MAE). The GNN was able to capture complex relationships between nodes in the supply chain network, such as those between suppliers, manufacturers, and distributors. This allowed it to make more accurate predictions about future demands.
One of the key advantages of using a GNN is its ability to handle large amounts of data. In this study, the dataset contained over 1 million products and their corresponding supply chains. The GNN was able to process this data efficiently and accurately, making it well-suited for real-world applications.
Another advantage of using a GNN is its ability to capture non-linear relationships between nodes in the supply chain network. Traditional methods often rely on linear models that fail to capture these complex relationships. However, the GNN’s architecture allows it to learn non-linear patterns and make more accurate predictions as a result.
The study also highlights the importance of using domain-specific knowledge when training a GNN. The researchers used their expertise in supply chain management to design the model and select the most relevant features from the dataset. This allowed them to improve the accuracy of the model and better capture the complex relationships between nodes in the supply chain network.
In addition to its ability to handle large amounts of data and capture non-linear relationships, the GNN also has the potential to be used for other tasks in supply chain management. For example, it could be used to predict delivery times or optimize inventory levels.
Cite this article: “Graph Neural Networks Improve Demand Forecasting in Supply Chain Management”, The Science Archive, 2025.
Supply Chain Management, Demand Forecasting, Graph Neural Networks, Gnns, Multi-Layer Perceptron, Mlp, Graph Convolutional Network, Gcn, Mean Squared Error, Mse, Mean Absolute Error, Mae.
