Detecting Vessel Performance Degradation Using XGBoost and LightGBM Driven by Explainable AI (SHAP)

Hibban, Muhammad Alif (2026) Detecting Vessel Performance Degradation Using XGBoost and LightGBM Driven by Explainable AI (SHAP). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5010221049-Undergraduate_Thesis.pdf Restricted to Repository staff only Download (10MB) | Request a copy

Abstract

Fuel oil consumption constitutes 50-60% of total maritime operational costs with hull biofouling inducing efficiency loses of up to 40%. Traditional fixed-interval dry-docking is economically inefficient as it fails to account for variable fouling rates. This research develops a data-driven vessel performance monitoring and degradation detection framework for Shipping Company using XGBoost and LightGBM models trained on operational noon report data. To ensure model transparency in high-stakes environments, SHapley Additive exPlanations (SHAP) were integrated to quantify feature contributions. Evaluation results demonstrated that LightGBM outperformed XGBoost, achieving an R2 of 0.974, a Mean Absolute Error (MAE) of 9.3798 Liters/Hour, a MAPE (Mean Absolute Percentage Error) of 3.31%, and a 3 ms processing time. SHAP analysis has identified RPM as the primary driver of consumption with an absolute mean SHAP value of 141.72. Model validation through Shapiro-Wilk normality test and a subsequent Mann-Whitney U-Test confirmed a statistical difference (p < 0.05) between pre-docking and post-docking residuals which proves the model’s ability to distinguish between fouled and clean hull states. This framework provides a deployable, data-driven decision support system to optimize dry-docking schedules and minimizing unnecessary fuel expenditure.
=================================================================================================================================
Biaya bahan bakar menyumbang 50-60% dari total biaya operasional maritim di mana pengotoran biologis lambung kapal memicu inefisiensi hingga 40%. Metode dry-docking terjadwal secara tradisional terbukti tidak efisien secara ekonomi karena mengabaikan variasi laju akumulasi pengotoran. Penelitian ini mengembangkan kerangka analisis kinerja kapal berbasis data untuk Shipping Company menggunakan algoritma XGBoost dan LightGBM berbasis data noon report operasional. Untuk menjamin transparansi model pada lingkungan dengan risiko tinggi, integrasi SHapley Additive exPlanations (SHAP) diterapkan untuk mengukur kontribusi fitur secara kuantitatif. Hasil evaluasi menunjukkan bahwa LightGBM unggul daripada XGBoost dengan mencapai R2 sebesar 0.974, Mean Absolute Error (MAE) 9.3798 Liter/Jam, Mean Absolute Percentage Error (MAPE) 3.31%, dan waktu pemrosesan 3 ms. Analisis SHAP mengidentifikasi RPM sebagai pendorong utama konsumsi bahan bakar dengan nilai rata-rata absolute SHAP sebesar 141.72. Validasi model melalui uji normalitas Shapiro-Wilk serta uji Mann-Whitney U mengonfirmasi perbedaan signifikan secara statistik (p < 0.05) antara residu sebelum dan sesudah docking yang membuktikan kemampuan model dalam membedakan kondisi lambung kotor dan bersih. Kerangka kerja ini menyediakan sistem pendukung keputusan berbasis data yang siap pakai untuk mengoptimalkan jadwal dry-dock dan meminimalkan pemborosan bahan bakar.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Vessel Performance, Hull Biofouling, Machine Learning, Explainable AI, Kinerja Kapal, Pengotoran Lambung, Machine Learning, Explainable AI.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA336 Artificial Intelligence V Naval Science > VK > VK361 Mooring of ships. Dry docks V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM163 Hulls (Naval architecture) V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM751 Resistance and propulsion of ships
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Industrial Engineering > 26201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Alif Hibban
Date Deposited:	20 Jul 2026 06:39
Last Modified:	20 Jul 2026 06:39
URI:	http://repository.its.ac.id/id/eprint/135322

Actions (login required)

View Item