Time Series Analysis and Prediction Statistical Models for the Duration of the Ship Handling at an Oil Terminal

Abstract

The main points of this paper are researching of time series and, then, building statistical prediction models based on obtained characteristics. Such investigation is often conducted in medicine and economics. The practice shows, that the studying of such issues in the marine transport logistics area is in demand. Time series analysis comprises methods for analyzing time series data in order to extract meaningful characteristics of the data and forecast future, based on knowledge of the past. The following models are applied to ship handling duration data: exponential smoothing (single, double and triple), and two more sophisticated models – auto-regressive moving average (ARMA) and ARIMA with seasonal component (SARIMA). The choice of a suitable model depends both on the distribution of the data and on the useful information that it will bring. Experiments for every model are carried out with different values of the model parameters to find the best fitting one. For all models Mean Square Error (MSE) is calculated on the training and test data. The best results were achieved by ARMA model with weekly dataset frequency. Nevertheless, there are some other ways to improve prediction models and to obtain more accurate results, which are also proposed in the article.