A navigation risk assessment model for small-sample multimodal data with an improved Bayesian network

In marine environmental risk scenarios, traditional Bayesian network models face challenges in achieving precise risk assessments due to insufficient data availability and linguistic uncertainties. To address this challenge, a small-sample multimodal model with improved Bayesian network (SSM-IBN) is proposed for navigational risk assessment of ocean-going vessels. The framework incorporates two key innovations: (1) A multidimensional Bootstrap expansion technique is developed to capture statistical feature from high-dimensional small-sample datasets, addressing the poor performance of Bayesian networks under limited data conditions. (2) Hesitant fuzzy agglomerative hierarchical clustering algorithm is proposed to handle subjective data uncertainties that conventional Bayesian approaches cannot adequately process. Furthermore, systematic validation is conducted through simulation experiments utilizing four distinct training sample sets, followed by the practical application of the model to risk analysis in North Pacific Ocean shipping. Experimental results show that, compared with traditional Bayesian networks, the SSM-IBN achieves an 18.2%–58.1% reduction in RMSE and a 0.294–1.0185 increase in R 2 when the expansion multiplier k = 9. Practical implementation in maritime risk assessment further confirms the model's advantages in small-sample scenarios through comprehensive visual analytics. • A small-sample multimodal model with improved Bayesian network is proposed for navigational risk assessment. • A multidimensional Bootstrap expansion is developed to capture statistical features from small samples. • Hesitant fuzzy agglomerative hierarchical clustering is proposed to handle subjective uncertainties.