ANALISIS DAN EVALUASI METODE MACHINE LEARNING UNTUK PREDICTIVE MAINTENANCE MESIN KAPAL: STUDI PERBANDINGAN SVM, GRADIENT BOOSTING, DAN LSTM

Arya Laksamana Jalma, Dr. Nur Rokhman, S.Si., M.Kom.

2026 | Skripsi | ILMU KOMPUTER

Industri maritim merupakan salah satu sektor vital yang menopang perekonomian global modern, di mana sekitar 80% volume perdagangan dunia bergantung pada transportasi laut, sehingga keandalan mesin kapal menjadi faktor penting dalam menjaga kelancaran operasional (Kalafatelis et al., 2024). Kegagalan mesin secara mendadak dapat meningkatkan biaya perawatan dan mengganggu operasional kapal, sehingga diperlukan pendekatan predictive maintenance berbasis data untuk mendeteksi kondisi mesin lebih awal. Penelitian ini bertujuan menganalisis dan membandingkan performa algoritma Support Vector Machine (SVM), Gradient Boosting, dan Long Short-Term Memory (LSTM) dalam klasifikasi kondisi mesin diesel kapal ke dalam tiga kelas, yaitu Normal, Requires Maintenance, dan Critical. Dataset sekunder dari Kaggle digunakan dengan parameter operasional mesin berbentuk time series, kemudian melalui tahapan preprocessing, pembagian data, dan hyperparameter tuning pada masing-masing model.

Hasil evaluasi menggunakan accuracy, precision, recall, F1-score, dan confusion matrix menunjukkan bahwa model Gradient Boosting menghasilkan performa terbaik dengan accuracy sebesar 34.62%, precision sebesar 34.48%, recall sebesar 34.55%, dan F1-score sebesar 34.18%, diikuti oleh model LSTM dan SVM. Hasil penelitian menunjukkan bahwa preprocessing dan hyperparameter tuning berpengaruh terhadap peningkatan performa model, meskipun klasifikasi antar kondisi mesin masih menghadapi tantangan akibat kemiripan pola data. Dengan demikian, Gradient Boosting dinilai paling optimal untuk implementasi predictive maintenance mesin kapal dan berpotensi mendukung peningkatan efisiensi perawatan serta keandalan operasional industri maritim.

The maritime industry is one of the vital sectorss supporting the modern global economy, with approximately 80% of world trade volume relying on sea transportation, making ship engine reliability a crucial factor in maintaining operational continuity (Kalafatelis et al., 2024). Unexpected engine failures may increase maintenance costs and disrupt ship operations, creating the need for data-driven predictive maintenance appproaches to detect engine conditions at an earlier stage. This study aims to analyze and compare the performance of Support Vector Machine  (SVM), Gradient Boosting, and Long Short-Term Memory (LSTM) algorithms in classifying ship diesel engine conditions into three categories: Normal, Requires Maintenance, and Critical. Secondary data obtained from Kaggle consisting of engine operational time-series parameters were used, followed by preprocessing, data splitting, and hyperparameter tuning for each model.

Evaluation results using accuracy, precision, recall, F1-score, and confusion matrix indicate that the Gradient Boosting model achieved the best performance with an accuracy of 34.62%, precision of 34.48%, recall of 34.55%, and F1-score of 34.18%, followed by the LSTM and SVM models. The findings demonstrate that preprocessing and hyperparameter tuning contributed to improving model performance, although classification among engines conditions remained challenging due to similarities in data patterns. Therefore, Gradient Boosting is considered the most suitable model for predictive maintenance of ship engines and has the potential to improve maintenance efficiency and operational reliability in the maritime industry.

Kata Kunci : Predictive Maintenance, Machine Learning, SVM, Gradient Boosting, LSTM, Marine Engine.