Navigating non-traditional risk: a deep residual networks framework on assessing piracy incident severity

Existing studies have predominantly focused on predicting the probability of pirate incidents, while lacking systematic analysis of consequence severity and comprehensive identification of the key factors contributing to more severe incidents. This study proposes an innovative framework based on Deep Residual Networks (ResNet) to assess the severity of piracy incidents and to uncover the complex relationships influencing the severity. A comprehensive piracy incident influential factors index is developed considering both the hard indicators including the ship’s features, external assistance, pirate’s ability, and the soft indicators including the corruption conditions, economy, and political stability. By addressing issues of missing data, imbalanced datasets and data noise through the application of the matrix completion algorithm, Adaptive Synthetic Sampling Approach and soft threshold function, we have compiled a robust database of piracy incidents in the Southeast Asian maritime region. The findings indicate that our proposed framework achieved an accuracy of 0.94, representing improvements of 1% and 2% over the Random Forest and Decision Tree models, respectively. Furthermore, variables such as seasonality, timing, weaponry employed by pirates, and vessel type significantly affect the severity of piracy incidents. This study provides a detailed decision-making foundation for stakeholders to formulate targeted anti-piracy strategies and policies.