Customer Segmentation of Port Based on the Multi-instance Kernel K-aggregate Clustering Algorithm

The analyses of the port data show us that the traditional data lay-out and the exited clustering algorithms could not be used in the port customer segmentation, so this thesis presents a new three-level data bag by combined with the way in which the multi-instance learning treat the data. Then a multi-instance kernel function is constructed according to the new bag. When the distance between two mixed valued vectors is counted the information gains are imported to weight the different attributes. The partition coefficient and average fuzzy entropy are calculated to decide the best cluster number of the clustering algorithm. Finally the kernel k-aggregate clustering algorithm using the multi-instance kernel is applied to the customer segmentation and gets a good clustering result which provides the managers guidance and evidence of different marketing strategies for corresponding subdivided markets.