Fault Diagnosis of Marine Engine Systems Using Optimized XGBoost with Grid Search Hyperparameter Tuning

The reliability of the main engine is a critical factor in ensuring vessel safety and operational effectiveness, particularly in the maritime industry, which requires continuous transportation services. Conventional maintenance approaches, such as time-based maintenance and corrective maintenance, remain widely applied; however, they often fail to provide early detection of potential failures. Disturbances such as overheating, lubrication degradation, fuel contamination, and turbocharger malfunction can significantly reduce engine performance and may result in unexpected operational shutdowns. Therefore, data-driven approaches based on machine learning are needed to improve monitoring capability and prediction of engine conditions. This study develops a ship engine condition classification model using the XGBoost algorithm based on multivariate data. The model training was conducted using five hyperparameter configurations with three train–test data split schemes (0.2, 0.3, and 0.5). Experimental results demonstrate stable performance, with accuracy ranging from 0.843 to 0.850, and improved performance as the proportion of training data increased. Models incorporating regularization showed better generalization capability when trained on larger datasets. Furthermore, hyperparameter optimization using Grid Search produced the optimal configuration for the final model. The findings indicate that XGBoost effectively handles complex sensor data and has strong potential to support reliable early fault detection in ship engines