Shipping companies are looking to minimize losses or maximize profits by accurately predicting the direction and the magnitude of the fluctuations in a constantly changing maritime situation. Therefore, in order to predict the maritime market more precisely, economic model between various variables such as demand, supply and freight rate etc. of shipping service has been established and forecasted. However, the determinants of the maritime markets are very diverse and volatile, and the decision mechanism is complex. The accurate prediction of the direction and the magnitude of the variation remains as a difficult challenge.
So, the purpose of this study is to propose an optimal Artificial Neural Network (ANN) model for dirty tanker markets forecasting through VLCC, SUEZMAX and AFRAMAX tanker market prediction using ANN. The data used in this ANN forecasting are 204 monthly time series data from 2000 to 2016. The ANN training algorithm was applied in two methods, Levenberg-Marquardt algorithm and Bayesian regularization algorithm, to forecast the tanker markets with the multi-step advanced time of one month, 3 months, 6 months, 9 months, 12 months and 15 months. And the performance accuracy of each algorithm was compared with.
In addition, the hidden layer size and the test data size of the Neural Network structure were changed and the predicted results were compared and evaluated to find an optimal ANN model for tanker market prediction. Furthermore, it was investigated the effect of the correlation between input and target variables on the ANN prediction when the size of each input variables change, and ANN forecasts were performed for three types of VLCC, SUEZMAX, AFRAMAX of dirty tankers which have different market fluctuations, and evaluated the accuracy and propriety of the results.
As a result of the study, the predictions results for VLCC, SUEZMAX and AFRAMAX tanker by Bayesian regularization algorithm are more satisfactory than those predicted by the Levenberg-Marquardt algorithm. In the one month, 3 months, 6 months and 9 months ahead predictions, the ANN structure with less number of hidden layer neurons than the number of input variables is more satisfactory than the structure with a larger number of hidden layer neurons. In the 12 months and 15 months ahead predictions, satisfactory results are obtained in the ANN structure with a larger number of hidden layers than the input variables, rather than the structure with fewer neurons than the input variables.
In the correlation between the target variable and the input variable, when the magnitude of the input variable has a strong correlation intensity with the target variable is changed, there is no significant change in the prediction performance error. However, when the size of the input variable with weak correlation strength is changed, the prediction performance error varies greatly. Predictions for the dirty tanker markets using ANN will help to minimize the risk of financial and operational problems.
Also, the forecasting information can be used as a very practical and effective means for establishing financial strategy and risk assessment. However, in order to use the prediction results as reliable information, it is most important to select the optimal artificial neural network model for the object to be predicted.