Ships are usually operated for more than 20 years, incurring large maintenance costs. Therefore, they require efficient and preemptive maintenance technology. Current research has focused on machine learning for predictive maintenance (PdM) of ships, but PdM data collection is limited because ship undergoes preventive maintenance and access to detailed data is restricted. Therefore, abnormal signs of ship equipment are not easily detected. To enable applications of machine-learning PdM to ship generator engines, this study collected and analyzed the data from operating ships. The abnormal data of the engine needed for machine learning were collected through engine simulations. In addition, generator engine condition criterion value (GCCV) was defined to determine anomalous symptoms based on the exhaust gas temperature by analyzing maintenance items. Next, a factor that corrects the GCCV under specific engine-operating conditions was derived using a regression analysis for establishing revision GCCV (RGCCV). The Mean Absolute Error (MAE) of RGCCV, which is calculated as a correction-factor, showed reliable results of 3.331–4.054 in all cylinders. After configuring and verifying the RGCCV-based engine-anomaly detection algorithm, anomalies during operation could be detected under normal conditions of the engine.