Logistics Management Optimization through Machine Learning: A Predictive Model for Item Transfer Time in Warehouse Activity-Space

Abstrak

Operational efficiency in warehouse logistics relies heavily on accurately predicting item transfer time. This study presents a machine learning-based framework using Gradient Boosting Classifier to classify transfer durations in the dynamic Jakarta Centrum warehouse, managed by the Corruption Eradication Commission (KPK) and PosIND. Field observations revealed inefficiencies due to unstructured layouts and fluctuating volumes. To improve prediction accuracy, the model incorporates Z-score normalization, SMOTE for class balancing, and hyperparameter tuning using GridSearchCV and PSO. The optimized model successfully classified 258 High, 285 Low, and 277 Medium transfer-time instances. SHAP analysis identified distance, distribution volume, and throughput as key influencing factors. Results demonstrate the potential of predictive modeling to enhance warehouse operations through better space usage, workforce planning, and SLA compliance. This study supports machine learning as a strategic tool for data-driven logistics optimization, with future work recommended to include contextual variables like workforce capacity and shift schedules for improved precision and real-world applicability.