Predicting fatigue life is essential in engineering applications for ensuring the safety and durability of structures and components under cyclic loading. This research introduces a novel machine learning method using Graph Convolutional Networks (GCN) to predict the fatigue life of L-PBF AlSi7Mg alloy. The novel aspect of this work is based on a representation of fatigue samples as graphs, each sample consisting of nodes. This graph-based structure enables a spatial defects analysis of samples, allowing for an in-depth investigation of how defect distribution contributes to the overall fatigue behavior of the alloy. Each node in the graph corresponds to a distinct region of the material and it includes a set of features such as the defect count per unit area, mean defect volume, average defect surface area and mean aspect ratio of defects. These features provide a multi-faceted understanding of the material’s defect distribution, which is crucial for identifying potential crack initiation sites and predicting how these imperfections will influence crack propagation under cyclic loading conditions. The GCN model offers a more accurate prediction of fatigue life compared to traditional methods, which often rely on simplified assumptions and less detailed analyses. The research demonstrates that GCNs can significantly enhance the accuracy of fatigue life prediction for L-PBF AlSi7Mg alloy in high and very high fatigue life regimes. The findings suggest that this method could contribute to the improved design of materials and structures subject to cyclic loadings, potentially leading to safer and more reliable engineering solutions. Future work will aim to incorporate additional node features such as microstructure feature(s) and test the model’s applicability to a broader range of materials by further expanding its predictive capabilities.