Result: Development of an Interpretable Maritime Accident Prediction System Using Machine Learning Techniques

Every year, maritime accidents cause significant losses to human life, vessels, and the environment, impacting over 80% of global trade by volume. These incidents not only disrupt global trade but also pose long-term ecological and economic risks. To address this challenge, this thesis proposes a Machine Learning–based Maritime accident prediction system that enables proactive risk assessment and prevention. Unlike conventional studies, which primarily conduct ex-post statistical analyses of accidents, this work adopts a holistic, proactive, and interpretable approach, integrating vessel, environmental, operational, and human factors into a unified predictive framework. The research begins with an extensive literature review and expert consultations to identify relevant risk factors. Two datasets — raw maritime incident logs and a refined, structured dataset were collected from publicly available sources, integrated, cleaned, and pre-processed for analysis. Exploratory data analysis (EDA) and feature engineering were conducted to capture complex patterns and dependencies, while variable selection and hot-spot detection techniques were applied to prioritize key predictors. A range of Machine Learning models were evaluated and compared. The results demonstrate that the proposed system achieves competitive prediction accuracy, with models such as logistic regression and KNN showing promising performance for accident risk classification. To ensure trust and adoption, Interpretable Machine Learning (IML) methods (LIME, SHAP) were incorporated, allowing the system to provide transparent explanations of risk predictions. The system was implemented using Python, Flask, and SQLite, with a web-based interface for user interaction. Evaluation results confirm its effectiveness in predicting Maritime accidents and highlight its potential for real-world applications, including risk management by maritime authorities, safer voyage planning, and integration into future autonomous navigation systems. This thesis contributes to ...