Prediction cubes
Bee-Chung Chen, Lei Chen, Yi Lin, Raghu Ramakrishnan · Very Large Data Bases · 2005
In this paper, we introduce a new family of tools for exploratory data analysis, called prediction cubes. As in standard OLAP data cubes, each cell in a prediction cube contains a value that summarizes the data belonging to that cell, and the granularity of cells can be changed via operations such as roll-up and drill-down. In contrast to data cubes, in which each cell value is computed by an aggregate function, e.g., SUM or AVG, each cell value in a prediction cube summarizes a predictive model trained on the data corresponding to that cell, and characterizes its decision behavior or predictiveness. In this paper, we propose and motivate prediction cubes, and show that they can be efficiently computed by exploiting the idea of model decomposition.