ANFIS model for assessing near-miss risk during tanker shipping voyages

Abstract

Adaptive neuro-fuzzy inference system (ANFIS) was applied to predict the risk of near-miss incidents during tanker shipping voyages. Firstly, near-miss incidents recorded by a global tanker shipping management company were analysed. Four variables-type of operation, vessel's location, on-board position, and harm potential were selected to train and predict the risk levels of near-miss incidents. The selected variables were found to be correlated with the observed frequency at three risk levels, namely low, medium and high. Gravity factor (GF) was calculated using the frequency of the categories in each variable and their associated risk levels. The calculated GF values and the risk levels of near-miss incidents were used as input values in the ANFIS model. Triangular, Trapezoidal and Gaussian membership functions were used. Subsequently, fuzzy logical theory and artificial neural networks were applied to train the data. Causal factors in terms of direct contributory factors, indirect contributory factors and root contributory factors to the near-miss incidents were analysed. Risk control measures were also proposed to improve safety during tanker shipping.