Safety culture modelling in maritime

Safety culture is a main character in a series of industrial accident investigations that have taken place since the accident take place in Chernobyl. It is widely accepted that organizations with strong safety culture are effective in preventing work accidents. Understanding and modeling the safety culture, where many variables are influential, can be complex and challenging. Techniques that used for solving complex and challenging problems are improving day by day. For this reason, artificial intelligence-based methods and statistical techniques were used together in this thesis. While multiple logistic regression (MLR) was chosen among statistical techniques, artificial neural networks (ANN) were preferred among artificial intelligence methods. The relation between a dependent and many independent variables with more than two categorized can be examined by MLR analysis. Meanwhile, ANN is a system which can imitate the human brain, learn like people in this way, interpret the learned information, and can generate results using the outputs. In this way, it seemed possible to use ANN and MLR in modeling the effect of safety culture which contains many subtitles, on accidents. Within the context of this thesis, evaluation and classification of 218 accidents and near misses based on subtitles of safety culture were performed with MLR analysis and ANN method. MLR correctly classified 83.5% of the accident results and found six subtitles meaningful for the model. ANN on the other hand, correctly classified 88.1% of the accident results and found all of the subtitles meaningful for the model.