| Abstract [eng] |
Big data plays a significant role in all industries. Big data is very useful in the transport sector. Combining a variety of data into one single dataset, there are possibility to model passengers and cargo flows more efficiently. Air transport is one of the most convenient transportation mode. This transport mode is very helpful to quickly reach the destination or carry perishable products all over the world. Air transport is the main way to access the distant markets around the world. Freight transport is very important to the transport sector, which contributes significantly to economic development. For these reasons, air transport is one of the most important transportation mode. However, the process does not run smoothly unless it is monitored and analysed. Most of the studies, modelling and forecasting the demand for passenger or freight transport, are carried out analysing single passenger or cargo flow time series. However, in this paper is searched the best model with hight forecasting accuracy using more than one time series. Analysing the air cargo flows to the United States, air cargo flows are modelled by flight origin country using airline time series. In order to obtain the best possible models, the dataset is divided into four groups by income levels-economies using flight origin country according to the World Bank methodology (high-income, upper-middle-income, lower-middle-income and low-income economies). Each group is clustered into smaller groups, this step is necessary to notice the differences in income levels-economic groups. Moreover, in this paper, one of the cluster analysis aim is to cluster times series based on similarity, so that time series in the same cluster are similar as possible. In this case, cluster analysis is also useful while creating vector models. The modelling is made in two ways – using vector autoregressive model and multiple linear regression. When preparing the vector autoregressive models, firstly reduction of data dimension are made using principal component analysis. The results reveal that vector autoregressive models better simulate historical values than predict future values. The best prediction results have been achieved by applying multiple linear regression with independent variables: the consumer price index, fuel prices, dummy variables and one moth time series lag. |
