Predicting types of human-related maritime accidents with explanations using selective ensemble learning and SHAP method

• DOI: 10.1016/j.heliyon.2024.e30046
• Publication Date: 2024-04-26T15:51:25Z
• Authors: He Lan; Shutian Wang; Wenfeng Zhang
• Keywords: Ensemble Learning

Maritime accidents frequently lead to severe property damage and casualties, an accurate reliable risk prediction model is necessary help maritime stakeholders assess the current situation. Therefore, present study proposes a hybrid methodology develop explainable for accident types. Based on advantages of selective ensemble learning method, this pioneers introduce two-stage selection aiming enhance predictive accuracy stability model. Then, SHAP (Shapley Additive Explanations) method integrated identify effective mapping associations seafarers' unsafe acts their factors with results. The results demonstrate that developed achieves good performance 87.50% F1-score 84.98%, which benefits in assessing type advance, so as make proactive intervention measures.