Is entropy an indicator of port traffic predictability? The evidence from Chinese ports

Historical time-series data of container traffic in ports are characterised by strong randomness that causes the time-series prediction performance to vary significantly. In this sense, the prediction accuracy differs as a function of exogenous variables related to port characteristics and the maritime shipping context. Thus, the measuring problem of a priori container traffic predictability (i.e., the ability of methods to predict port traffic variables) arises. Academia suggests that this predictability may be related to entropy as a measure of the information provided. This paper analyses the Sample Entropy (SampEn), as an indicator of the complexity represented by the container traffic time-series data, comparing it with the error metrics provided by a statistical prediction and an artificial intelligent method. The Autoregressive Integrated Moving Average (ARIMA) and Back Propagation (BP) Neural Network are used to obtain the prediction performance of container port traffic in a data set composed of 20 Chinese ports and 5 selected international ports. Unexpectedly, the results indicate that the correlation between the SampEn and the prediction performance of the time series is weak. The complex ecosystem of the container shipping sector and the different port characteristics are suggested as the cause of this poor correspondence. Our findings help to understand the relationship between entropy and the predictability of time-series data in a sector (i.e. port traffic), which has not been explored before.