Prediction in OLAP Data Cubes

Abstract

Online analytical processing (OLAP) provides tools to explore data cubes in order to extract the interesting information, it refers to techniques used to query, visualise and synthesise the multidimensional data. Nevertheless OLAP is limited on visualisation, structuring and exploring manually the data cubes. On the other side, data mining allows algorithms that offer automatic knowledge extraction, such as classification, explanation and prediction algorithms. However, OLAP is not capable of explaining and predicting events from existing data; therefore, it is possible to make a more efficient online analysis by coupling data mining and OLAP to allow the user to assist in this new task of knowledge extraction. In this paper, we will carry on within works achieved in this theme and we suggest to extend the abilities of OLAP to prediction (enhancing the OLAP abilities and techniques by introducing a predictive model based on a data mining algorithms). The model is calculated on the aggregated data, and prediction is done on detailed missing data. Our approach is based on regression trees and neural networks; it consists to predict facts having a missed measures value in the data cubes. The user will have in his disposition, a new platform called PredCube, that offers the possibility to query, visualise and synthesise the multidimensional data, and also to predict missing values in the data cube using three data mining methods, and evaluate the quality of the prediction by comparing the average error and the execution time given by each one.