Urban water distribution networks (WDNs) are critical lifeline infrastructures, yet their risk assessment is challenged by complex multi-factor interactions. This study proposes a hybrid framework that integrates the N-K model for coupling analysis with Random Forest (RF) for data-driven prediction, combined through Bayesian weighted fusion to address this challenge. A four-dimensional index system including human, equipment, environmental, and management factors is established to identify diverse risk sources. N-K model analysis of 540 accident cases demonstrates significant four-factor coupling effects, suggesting synergistic interactions lead to cumulative risk amplification. For RF modeling, 920 pipe segment data points are analyzed, and the model outperforms alternative approaches in key performance metrics, demonstrating superior capability in identifying leakage-prone pipes. The fused model classifies risks into three levels, facilitating differentiated operation and maintenance strategies. Results indicate that the framework effectively enhances risk assessment accuracy, providing a practical tool for refined management of WDNs.