<![CDATA[The maritime logistics industry plays a crucial role in ensuring the smooth flow of trade; however, it often faces discrepancies between the number of containers booked and the number actually loaded onto ships. These discrepancies can lead to operational inefficiencies, shipment delays, and additional costs for companies. PT XYZ, as a maritime logistics service provider, encounters similar challenges. Therefore, this study aims to analyze the factors causing container discrepancies and to develop a predictive system for estimating the actual number of containers as a decision-support tool. This research adopts data mining, machine learning, and ensemble learning approaches, focusing on the Random Forest Regressor and Extreme Gradient Boosting (XGBoost) algorithms combined through a Voting Regressor. Hyperparameter tuning using GridSearchCV is applied to improve the model’s ability to capture complex data patterns. A quantitative approach following the CRISP-DM framework is employed, including data exploration, cleaning, feature selection, modeling, and evaluation. The study utilizes historical container booking data from PT XYZ in 2023, consisting of more than 138,000 records. The results show that the Voting Regressor achieves the best performance with an R² value of 0.7874 and an MSE of 1.6282, supported by consistent RMSE and MAE metrics. The model is implemented in a Flask-based web application that enables practical container count prediction through Microsoft Excel file uploads. The implementation of this predictive system has the potential to help PT XYZ reduce loading discrepancies, minimize additional costs, and optimize logistics planning, while also contributing academically to the application of machine learning in the maritime logistics sector.]]>
