Paper 4

Modular Neural Networks for Extending OLAP to Prediction

Authors: Wiem Abdelbaki, Sadok Ben Yahia, and Riadh Ben Messaoud

Volume 21 (2015)

Abstract

On-line Analytical Processing (OLAP) represents a good applications package to explore and navigate into data cubes. Though, it is limited to exploratory tasks. It does not assist the decision maker in performing information investigation. Thus, various studies have been trying to extend OLAP to new capabilities by coupling it with data mining algorithms. Our current proposal stands within this trend. It has two major contributions. First, a Multi-perspectives Cube Exploration Framework (MCEF) is introduced. It is a generalized framework designed to assist the application of classical data mining algorithm on OLAP cubes. Second, a Neural Approach for Prediction over High-dimensional Cubes (NAPHC) is also introduced, which extends Modular Neural Networks (MNN)s architecture to multidimensional context of OLAP cubes, to predict nonexistent measures. A preprocessing stage is embedded in NAP-HC to assist it in facing up the challenges arising from the particularity of OLAP cubes. It consists of an OLAP oriented cube exploration strategy coupled with a dimensions reduction step that reposes on the Principal Component Analysis (PCA). Carried out experiments highlight the efficiency of MCEF in assisting the application of MNNs on OLAP cubes and the high predictive capabilities of NAP-HC.