Analysis of merchant vessel accidents in Istanbul strait through decision tree method

Abstract

In this study merchant vessel accidents which occured between 2001 and 2016 in the sectors of Türkeli, Kandilli, Kadıköy, Marmara which constitutes Istanbul Strait region under Istanbul Vessel Traffic Services scope have been examined. Data was obtained from database of Ministry of Transport and Infrastructure Main Search and Rescue Coordination Center, and after data cleansing process, 535 vessel accidents which involve merchant cargo vessels of above 500 grosston have been analyzed. Merchant cargo vessel accidents which were taken place in the specified sectors have been examined with CHAID (Chi-square Automatic Interaction Detector) Decision Tree method. CHAID Decision Tree method is one of the most common used decision tree algorithms in extracting meaningful rules from big datasets and for classification. Through conducting CHAID Decision Tree method for merchant vessel accidents relationship between accident type (collision/contact, grounding and other) and vessel factors (vessel type, Length overall (LOA), vessel gross tonnage, vessel age, flag, loading condition), time factors (accident time, season of accident) and other factors (sector where accident occured, pilot on board or not) has been analyzed. Accident occuring sector, pilot on board/not, vessel type and accident time have been found as the most important input variables. Based on the result of the Decision Tree method applied to the data set, it was observed that the accidents occurring in the Kadıköy sector were collision / contact with 86% probability, the accidents occurring in the Kandilli or Marmara sectors were collision / contact with 48% probability and in the Türkeli sector, both collision / contact and other accident types had 36% occurring probability.