A method for identifying behavioral feature of risk navigation preference

Currently, research on the causes of improper maneuvering by ship drivers involves a lot of subjective judgments, resulting in significant deviations between research results and actual situations. Based on this, this article constructs a risk navigation preference behavior feature recognition model based on the Shapley Additive Interpretation (SHAP) algorithm and Extreme Gradient Boosting Tree (XG boost) technology. Actual data verification shows that risk navigation preference behavior can be divided into cautious, normal, and aggressive types, among which aggressive ships tend to frequently change their speed during navigation; The recognition accuracy of the model for three types of risk navigation preference behaviors reached 99.4%. Using the selected 11 key features as input variables for the model, the prediction accuracy still reached 99.2%. Although slightly lower than the former, the computation time decreased from 57.22ms to 24.15ms.