Determining the most influential human factors in maritime accidents: A data-driven approach. (1st September 2020)

• Record Type: Journal Article
• Authors: Coraddu, Andrea; Oneto, Luca; Navas de Maya, Beatriz; Kurt, Rafet
• Is Part Of: Ocean engineering. Volume 211(2020)
• Publication Date: 2020-09-01
• Subjects: Human factors -- Shipping accidents -- Accident investigation -- Data analytics -- Random forests -- Kernel methods -- Boolean kernels -- Feature ranking -- Gini impurity -- Backward elimination; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2020.107588
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: Marine accidents are complex processes in which many factors are involved and contribute to accident development. For this reason, effectively analyse what combination of factors lead an accident event is a complex problem, especially when human factors are involved. State-of-the-art methods such as Human Factor Analysis and Classification System, Human reliability Assessments, and simple Statistical Analysis are not effective in many situations since they require the intervention of human experts with their limitations, biases, and high costs. The authors propose to use a data-driven approach able to utilise the information present in historical databases of marine accident for the purposes of establishing the most influential human factors. For this purpose a two-stage approach is presented: first, a data-driven predictive model is built able to predict the type of accident based on the contributing factors, and then the different contributing factors are ranked based on their ability to influence the prediction. Results on a real historical database of accidents provided by the Marine Accident Investigation Branch, an independent unit within the UK Department for Transport, will support the proposed novel approach. Highlights: A data-driven model that predict the accident type based on the contributing factors. A kernel-based model to find the most influential human factors in maritime accident. Results on real data proved high accuracies in predictions. A tool for researchers and policymakers to identify the causes of maritime accidents.