Pemodelan Peluang Tubrukan Kapal dengan Pendekatan Machine Learning Menggunakan Copula Bayesian Network

Ratih, Iis Dewi (2025) Pemodelan Peluang Tubrukan Kapal dengan Pendekatan Machine Learning Menggunakan Copula Bayesian Network. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7019221002-Dissertation_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Aktivitas pelayaran di perairan Indonesia menunjukkan tren peningkatan yang signifikan, ditandai oleh bertambahnya jumlah kapal yang melintas serta pembangunan instalasi laut seperti platform, pipa, dan kabel bawah laut. Kondisi ini turut meningkatkan potensi terjadinya kecelakaan, khususnya tabrakan antar kapal. Berdasarkan data dari Direktorat Jenderal Perhubungan Laut dan Direktorat Kesatuan Penjagaan Laut dan Pantai, tercatat 1.601 kecelakaan kapal terjadi selama periode 2007–2018, di mana 11% di antaranya berupa tabrakan. Sementara itu, laporan dari Komite Nasional Keselamatan Transportasi (KNKT) mencatat 209 kasus kecelakaan kapal sepanjang 2007–2022, dengan proporsi tabrakan mencapai 20%. Temuan ini menunjukkan bahwa tabrakan merupakan jenis kecelakaan laut yang paling sering diinvestigasi KNKT, sehingga menjadi fokus utama dalam upaya mitigasi risiko pelayaran. Tabrakan kapal dapat dipicu oleh berbagai faktor, antara lain kondisi alam, kegagalan teknis, keterbatasan sumber daya, kesalahan navigasi, human error, serta kelalaian dari pihak kapal lain. Untuk menangani kompleksitas interaksi antar penyebab yang bersifat non-linier, diperlukan suatu model prediktif yang tidak hanya memetakan peluang kecelakaan, tetapi juga mampu menangkap hubungan kausal yang saling bergantung. Penelitian ini mengembangkan pendekatan Copula Bayesian Network (CBN) guna memodelkan causation probability (Pc) tabrakan kapal dengan mempertimbangkan dependensi linear maupun non-linear antar variabel, serta mengintegrasikan data diskrit dan kontinu dalam satu kerangka analisis. Data yang dianalisis meliputi kasus tabrakan kapal yang diinvestigasi oleh KNKT dan Mahkamah Pelayaran dalam kurun 2007–2022, dengan fokus pada tiga skenario utama: head-on, crossing, dan overtaking. Model yang dikembangkan meliputi Bayesian Network (BN), Hybrid BN, dan Copula BN (CBN), yang dibandingkan berdasarkan confusion matrix dan akurasi prediksi. Hasil evaluasi menunjukkan bahwa model BN memberikan estimasi Pc yang paling sesuai untuk masing-masing skenario, yaitu: head-on sebesar 2,13 × 10−4, crossing sebesar 1,58 × 10−4, dan overtaking sebesar 3,58 × 10−5. Namun, dalam mengidentifikasi faktor penyebab secara mendalam, model CBN terbukti lebih sensitif dalam menilai kontribusi setiap variabel terhadap kemungkinan tabrakan. Melalui analisis sensitivitas, faktor-faktor utama yang paling berkontribusi terhadap insiden tabrakan meliputi: (1) petugas jaga tidak menjalankan tugas sesuai prinsip good seamanship, (2) pengambilan keputusan yang buruk, dan (3) tindakan pencegahan yang tidak tepat. Kontribusi utama dari penelitian ini adalah pengembangan model CBN yang lebih adaptif terhadap kompleksitas sistem pelayaran di Indonesia, dibandingkan pendekatan konvensional atau default IWRAP yang menggunakan Pc nilai dari luar negeri. Model ini diharapkan dapat meningkatkan akurasi estimasi frekuensi tabrakan kapal serta mendukung penyusunan kebijakan keselamatan pelayaran yang kontekstual, berbasis data, dan berbasis risiko.
==================================================================================================================================
Maritime activities in Indonesian waters have shown a significant upward trend, marked by the increasing number of vessels and the expansion of marine installations such as platforms, pipelines, and underwater cables. This situation has also amplified the potential for accidents, particularly ship collisions. According to data from the Directorate General of Sea Transportation and the Sea and Coast Guard Unit, there were 1,601 ship accidents reported between 2007 and 2018, with 11% involving collisions. Meanwhile, the National Transportation Safety Committee (KNKT) recorded 209 ship accident cases from 2007 to 2022, with collisions comprising 20%. These findings indicate that collisions represent the most frequently investigated type of maritime accident by KNKT and are therefore a critical focus in risk mitigation efforts. Ship collisions can be triggered by various factors, including natural conditions, technical failures, resource limitations, navigational errors, human error, and negligence by other vessels. To address the complexity of nonlinear interactions among these causal factors, a predictive model is required that not only estimates accident probabilities but also captures interdependent causal relationships. This study introduces the Copula Bayesian Network (CBN) approach to model the causation probability (Pc) of ship collisions by accounting for both linear and nonlinear dependencies among variables, while integrating discrete and continuous data within a unified analytical framework. The dataset comprises ship collision cases investigated by KNKT and the Maritime Court between 2007 and 2022, focusing on three primary scenarios: head-on, crossing, and overtaking. The models developed include the Bayesian Network (BN), Hybrid BN, and Copula BN (CBN), which were evaluated and compared using confusion matrix and prediction accuracy. The results indicate that the BN model provides the most appropriate Pc estimates for each scenario—head-on: 2.13 × 10−4, crossing: 1.58 × 10−4, and overtaking: 3.58 × 10−5. However, for a more in-depth analysis of causal factors, the CBN model demonstrates greater sensitivity in identifying each variable's contribution to collision likelihood. Sensitivity analysis reveals the most significant contributing factors to ship collisions include: (1) watchkeeping officers failing to perform duties according to good seamanship principles, (2) poor decision-making, and (3) inappropriate preventive actions. The main contribution of this study lies in the development of a CBN model that is more adaptive to the complexities of Indonesia's maritime navigation system compared to conventional or default approaches such as IWRAP, which apply generic Pc values from foreign contexts. This model is expected to enhance the accuracy of ship collision frequency estimation and support the formulation of safety policies that are contextual, data-driven, and risk-based.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Kata kunci: Analisis Risiko, Bayesian Network, Causation Probability, Copula, Keselamatan Pelayaran, machine learning, Tubrukan Kapal Keywords: Risk Analysis, Bayesian Network, Causation Probability, Copula, Maritime Safety, Machine Learning, Ship Collision
Subjects:	Q Science > QA Mathematics > QA279.5 Bayesian statistical decision theory. V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM293 Shipping--Indonesia--Safety measures V Naval Science > VM Naval architecture. Shipbuilding. Marine engineering > VM731 Marine Engines
Divisions:	Faculty of Marine Technology (MARTECH) > Marine Engineering > 36001-(S3) PhD Theses
Depositing User:	Iis Dewi Ratih
Date Deposited:	07 Aug 2025 01:26
Last Modified:	07 Aug 2025 01:31
URI:	http://repository.its.ac.id/id/eprint/127890

Actions (login required)

View Item