A Neural-Based Approach for Extending OLAP to Prediction

In the Data Warehouse (DW) technology, On-line Analytical Processing (OLAP) is a good applications package that empowers decision makers to explore and navigate into a multidimensional structure of precomputed measures, which is referred to as a Data Cube. Though, OLAP is poorly equipped for forecasting and predicting empty measures of data cubes. Usually, empty measures translate inexistent facts in the DW and in most cases are a source of frustration for enterprise managements, especially when strategic decisions need to be taken. In the recent years, various studies have tried to add prediction capabilities to OLAP applications. For this purpose, generally, Data Mining and Machine Learning methods have been widely used to predict new measures' values in DWs. In this paper, we introduce a novel approach attempting to extend OLAP to a prediction application. Our approach operates in two main stages. The first one is a preprocessing one that makes use of the Principal Component Analysis (PCA) to reduce the dimensionality of the data cube and then generates ad hoc training sets. The second stage proposes a novel OLAP oriented architecture of Multilayer Perceptron Networks (MLP) that learns from each training set and comes out with predicted measures of inexistent facts. Carried out experiments demonstrate the effectiveness of our proposal and the performance of its predictive capabilities.