<p>Water inrush are among the primary geological hazards encountered during tunnel construction. Therefore, it is essential to develop a scientific and effective risk assessment model for such hazards. However, existing methods suffer from poor generalizability, heavy reliance on subjective experience, and insufficient integration of multi-source data, making it difficult to comprehensively and accurately evaluate the risk of water and mud inrush. In this study, a novel risk assessment model for water inrush was proposed, which integrated knowledge graph-guided multi-source data fusion with machine learning. A water inrush case-based knowledge graph and knowledge graph embeddings were employed to achieve a unified representation of multi-source data. The embeddings extracted from the graph were used as features to construct risk assessment model. To validate the model, a comprehensive dataset of global water and mud inrush cases was collected to establish a karst tunnel disaster database. The model’s performance was evaluated using tenfold cross-validation and a separate test set, achieving an prediction accuracy of 97.67%. Furthermore, the model was validated in a real-world risk assessment for the Qiyueshan Tunnel, with results highly consistent with the on-site situation. The findings of this study provide valuable guidance for risk assessment of water inrush in tunnels.</p>