Title: Decision making on sea: an expert system for risk assessment in maritime using data mining
Authors: Dimitrios Kokotos; Alkiviadis Kyriakakis
Addresses: Department of Maritime Studies, University of Piraeus, Greece ' Alliance Manchester Business School, University of Manchester, UK
Abstract: This work proposes the prototyping implementation of a dynamic expert system. The essence is the proposal is prediction of ship accidents. The validation process is based on data collected from coast guard official investigation reports. A classifier based on C5 algorithm is able to work even in presence of limitations for real-world data (noisy, many missing attribute values, etc). C5 algorithm is used for building decision trees and the models are used in the knowledge acquisition and its representation. The optimal decision rules estimated the dependency of the most important predictor upon the target variable 'source of accidents'. The comparison between two time periods shows that accidents due to human error were reduced, a result in line with the IMO report. The resulting patterns can be used to gain insight into aspects of shipping safety and to predict outcomes for future situations as an aid to decision making.
Keywords: classification algorithms; prediction; ship accident; maritime safety; decision trees; data mining; off-shore.
DOI: 10.1504/IJIDS.2021.119372
International Journal of Information and Decision Sciences, 2021 Vol.13 No.4, pp.365 - 380
Received: 21 Aug 2018
Accepted: 16 Sep 2019
Published online: 02 Dec 2021 *