A novel feature engineering method for severity prediction of marine accidents

Abstract

Predicting the severity of marine accidents is a crucial issue in the field of maritime safety. To enhance the accuracy of predicting marine accident severity, this study proposes a three-stage feature engineering approach called Integrated Feature Engineering and Balancing (IFEB). In the first stage, a method called Association Rule Fusion is developed to simplify multiple interrelated risk influential factors into a composite factor. In the second stage, the Support Vector Machine Synthetic Minority Over-sampling Technique is employed to address the issue of class imbalance. In the third stage, a predictive model-based feature selection method is utilised to identify features that positively influence the predictive model. Subsequently, various advanced machine learning models are employed to predict the output of IFEB, and the optimal predictor is selected. Finally, a series of ablation studies are conducted to validate the contributions of each module within IFEB to the overall model performance. The research findings indicate that IFEB can improve the average performance of each predictive model by up to 6.43%, IFEB combined with Light Gradient Boosting Machine emerges as the most stable and effective predictive model. This study provides an effective predictive tool for enhancing maritime safety and reducing the risk of marine accidents.

Topics & Concepts