engleski

Combining Neural Networks and On-line Analytical Processing in Sale Prediction of Footware and Leather Goods

Methodological basis of a data warehouse system consists of classical statistical methods used with on-line analytical processing (OLAP) technology at the lower level, enhanced with more sophisticated data mining techniques at the higher level. The paper shows a way of integrating the two levels such that drill-up, drill-down, and drill-through results are used as the additional input in modeling neural networks. The research is performed on a real dataset collected from a Croatian footwear and leather goods company that deals with the problem of allocating their goods among a number of retail stores. Prediction of weekly sale for each of the stores is important for planning inventory and satisfying customer needs. OLAP drilling methods are used to reveal some key issues in previous sale on the first level of experiment. The results of these analyses are used to create new input variables included in the neural network model for sale prediction on the second level. The paper compares the results of the neural network model that use such additional knowledge with the model that use only raw data. Guidelines for future research regarding modeling in data mining systems are also discussed.

data mining; data warehousing; footwear and leather goods; neural networks; OLAP

nije evidentirano