To address the limitations of tunnel water inrush risk assessment caused by missing indicators and insufficient utilization of multi-source heterogeneous information under complex geological conditions, a comprehensive evaluation method integrating combined weighting and improved interval attribute identification theory was proposed. A multi-dimensional evaluation index system that incorporated geophysical exploration, monitoring, geological survey and drilling data was established. The weights of all indicators were determined using the entropy weight method and fuzzy analytic hierarchy process. Interval-valued intuitionistic fuzzy numbers were introduced to characterise indicator uncertainty, and an improved interval attribute recognition theory was employed for the risk assessment of water inrush hazards. For scenarios involving missing indicators, an enhanced Gaussian process regression model was employed to precisely evaluate small-sample time-series data. A Bayesian network was constructed based on mutual verification and complementarity relationships among indicators to handle discrete missing values through probabilistic inference, and a sensitivity analysis of missing values was conducted to validate the reliability of the water-inrush-hazard prediction results. Finally, The Second Subsea Tunnel of Jiaozhou Bay, Qingdao was selected as a case study for validation, and the assessment results were found to be consistent with actual field conditions. The results indicate that the proposed method integrates multi-source information and a missing-value compensation mechanism. The reliability and applicability of the assessment under missing-indicator scenarios are significantly enhanced. Consequently, a systematic theoretical tool and decision support are provided for dynamic risk management in complex geological tunnels.