Development of Spatial Clustering Method and Probabilistic Prediction Model for Maritime Accidents

Abstract

Maritime accidents are complex incidents characterized by the absence of physically defined routes, unlike roads, and influenced by various factors such as marine conditions, vessel status, and multiple contributing factors. Despite efforts to reduce maritime accidents in Republic of Korea, the number of accident occurrences remains at a high level. Moreover, the infrequency of severe accidents makes it difficult to gather sufficient data for comprehensive analysis. This study aims to construct and validate a maritime accident prediction model to protect life and property in the event of accidents, considering the characteristics of the ocean, where tracking and managing transportation means is challenging. For the first step of predicting maritime accidents, a 5-level spatial system was constructed based on the density of accidents and vessel traffic. The data to be used in the analysis was constructed through four steps. In order to evaluate the maritime accident prediction model, we conducted a comparative analysis with the machine learning-based ANN model using a Confusion matrix. Cost-of-omission analyses were conducted based on vessel traffic volume and environmental factors according to the seasons to examine the influence of variables on the probability of maritime accidents. Based on this, implications are presented through an analysis of the causes of maritime accidents according to their factors.