On the need of new approaches for the novel problem of long-term prediction over multi-dimensional data

Abstract

Mining evolving behavior over multi-dimensional structures is increasingly critical for planning tasks. On one hand, well-studied techniques to mine temporal structures are hardly applicable to multi-dimensional data. This is a result of the arbitrary-high temporal sparsity of these structures and of their attribute-multiplicity. On the other hand, multi-label classification over denormalized data do not consider temporal dependencies among attributes. This work reviews the problem of long-term classification over multidimensional structures to solve planning tasks. For this purpose, firstly, it presents an essential formalization and evaluation method for this novel problem. Finally, it extensively overviews potential relevant contributions from different research streams. © 2012 Springer-Verlag Berlin Heidelberg.