Predicting Vessel Departure Delays at Tanjung Pandan Port Using Supervised Machine Learning : A Comparative Study of Logistic Regression, Decision Tree, and SVM

Abstract

Operational delays in vessel departure disrupt maritime logistics and increase port dwell time. This study develops predictive models to anticipate departure delays at Tanjung Pandan Port using supervised machine learning. Three algorithms—Logistic Regression, Decision Tree, and Support Vector Machine (SVM)—were trained on 112 verified port calls (2023–2024) with key features: arrival date, scheduled departure date, vessel ownership status (milik vs. keagenan), and document response time. Delay was defined as exceeding the median turnaround time of 58 hours. Data preprocessing included imputation, time-difference engineering (e.g., ΔTIBA–BERANGKAT, response latency), and SMOTE for class balancing. Performance was evaluated using accuracy, precision, recall, and F1-score via 5-fold cross-validation. The Decision Tree model achieved the highest F1-score (0.86) and recall (0.89), identifying response latency > 12 hours, keagenan status, and arrival during neap tide windows as top predictors. SVM showed robust precision (0.88), while Logistic Regression offered the best interpretability of coefficient impact. The models collectively support proactive scheduling interventions-e.g., digital clearance acceleration or priority berthing for high-risk vessels—to mitigate delays. This study contributes the first ML-based delay prediction framework for shallow-draft, tramp-operated Indonesian ports.