engleski

Integration of OLAP and Neural Networks in Modeling Sale Prediction

According to other authors, neural networks outperform linear statistical methods in sale forecasting and prediction. For these reasons they were suggested as one of the data mining techniques in this domain, together with decision trees and statistical methods. The paper investigates the possibilities of integrative use of on-line analytical processing (OLAP) and neural networks in finding the best model for sale prediction of footwear and leather goods. A sale data cube is created using the real data from a Croatian company. Since our previous results showed that OLAP analyses and drilling methods are able to provide an additional knowledge into the neural network model, this research is focused on evaluating separate models that can be suggested by looking into the OLAP analyses. The accuracy of models is compared using the mean square error as the objective function. The sensitivity analysis is also performed in order to extract the most important variables for the sale prediction. The results can serve as guidelines for researchers as well as for practitioners that aim to use the best of both methodologies in improving the efficiency of their businesses.

neural networks; on-line analytical processing; data mining; sale prediction

nije evidentirano