A lag-enhanced machine learning framework for weather-induced ropax ferry risk prediction: a case study of the java-sumatra corridor

Pebrianti, Dwi and Setyawan, Sigit and Imelda, Imelda and Samad, Rosdiyana and Bayuaji, Luhur (2026) A lag-enhanced machine learning framework for weather-induced ropax ferry risk prediction: a case study of the java-sumatra corridor. In: 2026 10th International Conference on Mechatronics Engineering (ICOM), 8 September 2026, Kulliyyah of Engineering, IIUM.

PDF - Published Version Restricted to Repository staff only Download (1MB) | Request a copy

Abstract

Extreme convective weather poses a recurring threat to maritime transport along the Java-Sumatra corridor, one of Indonesia's busiest logistical routes. Ensuring the safety of RoPax ferry operations requires translating raw meteorological data into clear, tiered risk assessments. This study presents a lag-enhanced machine learning framework for weather-induced ferry risk prediction, classifying conditions into Safe, Warning, and Danger categories. The framework was developed using a 25-year hourly atmospheric dataset (2000–2025) containing 227,208 records of wind parameters (gust, speed, and direction), surface pressure, and precipitation. Because atmospheric changes tend to unfold gradually, the feature space was expanded with 6-hour temporal lags and trigonometric cyclic variables, allowing the model to account for short-term weather trends. To handle the class imbalance caused by the rarity of extreme storm events, the Synthetic Minority Over-sampling Technique (SMOTE) was applied within the training partition only. The proposed Random Forest model was evaluated using a chronological 80:20 validation split and 5-fold cross-validation, achieving an overall accuracy of 89% and a Matthews Correlation Coefficient (MCC) of 0.71, outperforming an XGBoost baseline on both metrics. For the high-priority "Danger" category, the model reached a precision of 0.78, a recall of 0.72, and an F1-score of 0.75, indicating a reasonable balance between detecting hazardous conditions and limiting false alarms. To demonstrate its practical value, the model is deployed as a proof-of-concept web prototype, giving port operators clear risk trends to ensure timely and confident safety decisions.

Item Type:	Proceeding Paper (Invited Papers)
Uncontrolled Keywords:	Early Warning System, Java-Sumatra Corridor, Machine Learning, Maritime Safety, Random Forest, RoPax Ferry, SMOTE, Temporal Lag.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T10.5 Communication of technical information T Technology > TC Hydraulic engineering. Ocean engineering T Technology > TC Hydraulic engineering. Ocean engineering > TC203 Harbors and coast protective works. Coastal engineering. Lighthouses T Technology > TK Electrical engineering. Electronics Nuclear engineering
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Engineering > Department of Mechanical Engineering
Depositing User:	Dr Dwi Pebrianti
Date Deposited:	02 Nov 2026 08:57
Last Update:	02 Nov 2026 08:57
Queue Number:	2026-10-Q5920
URI:	http://irep.iium.edu.my/id/eprint/132348
Indexed In:	Google Scholar

Actions (login required)

View Item