Predicting Maritime Crew Change Failure Using Explainable Machine Learning: A Risk-Based Early Warning Model for Seafarer Rotation Disruption

Crew change failure is a recurring operational risk in maritime shipping, yet most crew planning systems still react to disruption after the nominated replacement has already become unavailable. This study develops an explainable machine learning model for predicting crew change failure before the planned relief date. The model estimates the probability that a nominated replacement will fail to join the vessel as scheduled by integrating crew, vessel, port, documentation, travel, medical, and operational-delay variables. A structured dataset is designed for ۱,۲۰۰ historical crew change records, including ۲۳ predictive features and a binary failure label. Four predictive models are evaluated: logistic regression, random forest, XGBoost, and long short-term memory. Model performance is assessed using accuracy, precision, recall, F۱-score, ROC-AUC, and Brier score, with particular attention to recall because missing a high-risk crew change is more costly than issuing an early warning. The results show that XGBoost achieves the strongest predictive performance, with an accuracy of ۹۱.۴%, recall of ۸۸.۷%, F۱-score of ۰.۸۹۴, ROC-AUC of ۰.۹۴۶, and Brier score of ۰.۰۷۱. SHAP-based explainability identifies visa uncertainty, flight-connection complexity, port-call variability, short relief lead time, and expiring medical documentation as the most influential predictors of crew change failure. The findings provide shipping companies with an early warning mechanism for identifying fragile relief plans, prioritizing intervention, and reducing last-minute crew rotation disruption. The study contributes to maritime workforce analytics by shifting crew change management from reactive rescheduling toward predictive, risk-based disruption prevention.

کلیدواژه ها:

Maritime crew change ، Crew rotation disruption ، Explainable machine learning ، XGBoost ، SHAP ، Seafarer relief ، Predictive risk analytics ، Maritime workforce management