| Hi-index | 0.00 |
The purpose of this paper is to analyse the visualisation capability of self- organizing maps and their usability in distribution control. The self- organizing maps are a special form of neural networks that are good in clustering and dimensionality reduction and could therefore be used in observing the significant deviations between an actual and a norm delivery such as previous or expected deliveries. However, many factors, such as data pre-processing and the size of the map, affect the goodness of the map''s representation, and we do not have exact rules for creating an optimal map. Instead, we have methods to analyse the goodness of resulting maps and their visualisation. In practice this means that the building process includes a construction of many, even tens of maps and afterwards a selection of the best one for the purpose. However, we anticipate that if the network building is too complicated, the user will loose her or his interest to use it in distribution control. For this reason, we concentrate on two issues in this study. First, we illustrate the impact of the map size on the accuracy of the visualisation. Second, we investigate how the number of customers and their shipments in the training data set influences the deviation detection. As a result, we found that both issues affect the goodness of the map. However, when considering the maps'' usability, these are not too severe factors. Hence, we do not need to pay too much attention to them in network building for deviation detection in distribution control.