Machine Learning-Based Demand Forecasting for Frequently Replaced Spare Parts in Marine Engines

Reliable operation of marine engines requires accurate prediction of spare parts requirements. Conventional forecasting methods often fail to adequately capture the non-linear dynamics and complex patterns inherent in historical spare parts consumption sequences. To improve forecasting accuracy, this study compares conventional time series models - including moving average (MA) and ARIMA - with two machine learning models, Support Vector Regression (SVR) and Random Forest (RF), for monthly forecasting of spare parts demand associated with frequent failures of marine engines. A feature set with lagged consumption values and moving statistics was created to train and evaluate the model. Experimental results based on simulated data show that the random forest (RF) model outperforms the other models and achieves the lowest root mean square error (RMSE) and a significantly higher coefficient of determination (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>). This confirms the effectiveness of machine learning in predicting spare parts requirements for frequent failures of marine engines. The study provides a precise and reliable forecasting method for optimizing spare parts inventory and maintenance planning in the shipping industry.