Generation of equal length patterns from heterogeneous patterns for using in artificial neural networks

A challenging task is to classify Internet customers based on their heterogeneous search histories of shopping in the Internet. The problem is the data pattern itself. Each transition of a customer from one page to the next in purchasing a commodity is considered as an attribute and this is a pair of data. The purchase patterns consist of usually different length for different customers. We cannot classify customers using a neural network (NN) due to these two problems - pair of attribute and unequal lengths of data. Here, we have developed an algorithm that can automatically generate equal length data with non-pair attributes. Finally, we use an unsupervised competitive learning in order to classify them because we do not know how many classes are there. We found that most of the customers belong to single category or class. The results we obtained have a nice agreement with the customerpsilas goal. The goal of all customers is to reach a common target page and to purchase a commodity. Therefore, we can consider that they may belong to the same category or class.