استفاده از ابزارهای داده کاوی برای مطالعه ایمنی حمل و نقل در آبهای محدود شده

The effectiveness of the enforcement of the ISM-Code and the examination of its role in the distribution of causes of shipping accidents between human and non-human error was studied. All accidents involving Greek-flagged ships from 1995 to 2006, a time-scale which spans over the pre- and post-ISM period in navigational regions of restricted waters, were analyzed. The accident data was processed through a classification tree analysis which enabled the classification of various accident factors. The analysis revealed that although the human error maintained its position as the dominant factor in shipping accidents, there is also substantial evidence in support of the ISM-Code effective control over shipping accidents during the post-ISM period. The implementation of the ISM-Code led to an overall reduction of human-induced accidents in total. Furthermore, in terms of location, the ISM-Code improved the human-induced accident record within restricted waters. Conclusively, the ISM-Code constitutes an effective policy measure for shipping safety. The results of the classification tree analysis reported in the present work can be used by decision makers in companies and international organizations to build knowledge-based expert systems and augment their information in the field of safety policy and management.

The International Safety Management Code (ISM-Code) was enforced by the International Maritime Organisation (IMO) in 1998 in order to provide an international standard for the safe management and operation of ships and for the prevention of pollution. Accidents in shipping and marine environmental pollution are attributed to various factors. The UK Marine Accident Investigation Branch (MAIB) states that “one factor still dominates the majority of maritime accidents: human error” ( MAIB, 2000). In recognition of this statement, the ISM-Code aims at the promotion of safety in shipping by ensuring all onboard activities must adhere to an operating standard of “who does what and when”, according to the Safety Management System (SMS) which had to be maintained by all shipping companies after the implementation of the ISM-Code. To this extent, an analysis of shipping accidents through the application of classification trees structured on “human” or “non-human” split as the primary cause is a valuable tool in accessing the effectiveness of the ISM-Code. A lot of effort from researchers world-wide has been devoted in estimating dependency of shipping accidents from various factors (Celik et al., 2010, Grech et al., 2008, Tzannatos, 2005 and Tzannatos, 2002) as well as in assessing the effectiveness of the measure of the enforcement of the ISM-Code. Previous research in this area (Psaraftis et al., 1998) provides an ‘in-depth’ analysis of the human element as a factor in Greek shipping accidents during the 1984–1994 period and in conclusion it comments that “It is impossible to tell for each of the cases reviewed that the accident would not have occurred if ISM were in place for the ship in question. It is also early to assess the impact of ISM on the safety of the ships on which the Code has been implemented. This will take years to ascertain, and the analysis to do so will not be trivial. However, the very fact that ISM certification implies that all procedures related to the operation of the ship would at least be established, monitored and controlled, means that the risk of a situation getting out of hand would be minimised.” Though comparable data on the effectiveness of the ISM-Code are not as straightforward as they should be, firstly due to the presence of inconsistencies in reporting and administration between different Memoranda of Understanding (MoUs), as well as amongst the signatory nations of each MoU on the other hand due to the influence of subjective issues, such as the attitude of the crew, the ease of inspection, the inspector’s mood results have shown that the ISM system is working (Paris MoU CIC, 2007). However, the ISM-Code assessment should not only focus on providing answers if the ISM system has overall improved safety at sea, but also on providing answers to several detailed questions. Thus, the ISM-Code assessment procedure should reveal knowledge of ‘why people make mistakes’, or ‘in which cases of accidents ISM system has improved safety the most’ or ‘in which cases (if any) ISM has not improved safety’ and so on. The extraction of this type of information may lead to proposals of appropriate remedial action. Tools used for the ISM-Code assessment include mainly classical statistical analysis techniques (Giziakis et al., 1996) applied in the records of vessel detentions and deficiency notices produced within the framework of inspections performed by the relevant authorities. These techniques produce a thorough shipping accidents analysis giving the statistics of human influence upon shipping accidents. The study of shipping accidents with data mining techniques (Kokotos and Smirlis, 2005) may reveal information not already extracted by researchers who had used the classical statistical analysis techniques. A powerful data mining function is the classification tree, applied in order to produce optimal classification rules. This algorithm is trained to classify the cases of a dataset, as the dataset of shipping accidents used in the current work, to certain categories of a target (dependent) variable using the information of several predictive (independent or explanatory) variables. Hence, the resulting tree diagrams relate the categories of the target variable to its predictors (Goodman, 1979). Classification trees are easily understood by both experts and non-experts. The classification trees may be utilized in the context of a potential decision support system and a risk management information system that will record evaluate and process data for ship accidents. The current work aims at mining hidden information and estimating the dependency of the source of accidents (human or non-human) upon the year of accident, the location of accident, the vessel type, size and age. The mined information will be used to assess the role of the ISM-Code in the distribution of causes of shipping accidents between human and non-human error. Finally, the mined information will measure the effectiveness of the enforcement of the ISM-Code. The data mining was performed in accidents of Greek-flagged ships world-wide that cover both the pre-ISM-Code and post-ISM-Code periods. With respect to the choice of flag administration, it is assumed that Greek shipping, by virtue of its size and diversity, constitutes a valuable reference for the analysis of accidents.

The current work assessed the ISM-Code impact and measured the effectiveness of its enforcement. Our analysis indicated that human errors have been reduced after the introduction of the ISM-Code since a shipping accident after 1998 is 65.3% more likely to be attributed to “non-human error”. In addition, during the two periods (pre-ISM/post-ISM) the average annual change in the rate of accidents attributed to human error has improved by 49.3%. As it was extracted from the application of the data mining tool of the classification tree, the location of the accident is the most significant factor for predicting the human/non-human cause of a shipping accident. This finding reveals that navigational location continues to play an important role upon the shipping accidents. Along the branch of accidents in restricted area, the year of accident was found as the next important factor, hence the distinct influence of year of accidents upon the location of accidents. Furthermore, the classification tree analysis indicated a reduction of 28.1% of human errors within restricted waters for the post-ISM period compared to the pre-ISM period. This reduction is attributed to the implementation of the ISM-Code. Consequently, it is indicated that the ISM-Code constitutes an effective policy measure for shipping safety, particularly for human safety. The results of the classification tree analysis presented in this work build a knowledge-based expert system which can enable ship owners, ship managers, classification societies, underwriters, brokers and national authorities to provide valuable information for issuing risk assessment reports. It can also enable decision makers in companies and international organizations to augment their information in the field of safety policy and management. The classification tree may be utilized in the context of a potential decision support system and/or a risk management information system that will record, evaluate and process data for shipping accidents. Finally, the need for a continuous analysis of shipping accidents is dictated by the accumulated evidence for the predominance of the human factor and the ever increasing pressure for further improvement on the safety record of shipping. Although the ISM-Code implementation led to significant reduction of human-induced accidents, additional reduction is necessary. Further studies in this field will led to the adoption of additional regulations for the shipping safety.