Journal of Intelligent Information Systems

, Volume 40, Issue 3, pp 501–527

Unsupervised feature construction for improving data representation and semantics

  • Marian-Andrei Rizoiu
  • Julien Velcin
  • Stéphane Lallich
Article

DOI: 10.1007/s10844-013-0235-x

Cite this article as:
Rizoiu, MA., Velcin, J. & Lallich, S. J Intell Inf Syst (2013) 40: 501. doi:10.1007/s10844-013-0235-x

Abstract

Attribute-based format is the main data representation format used by machine learning algorithms. When the attributes do not properly describe the initial data, performance starts to degrade. Some algorithms address this problem by internally changing the representation space, but the newly constructed features rarely have any meaning. We seek to construct, in an unsupervised way, new attributes that are more appropriate for describing a given dataset and, at the same time, comprehensible for a human user. We propose two algorithms that construct the new attributes as conjunctions of the initial primitive attributes or their negations. The generated feature sets have reduced correlations between features and succeed in catching some of the hidden relations between individuals in a dataset. For example, a feature like \(sky \wedge \neg building \wedge panorama\) would be true for non-urban images and is more informative than simple features expressing the presence or the absence of an object. The notion of Pareto optimality is used to evaluate feature sets and to obtain a balance between total correlation and the complexity of the resulted feature set. Statistical hypothesis testing is employed in order to automatically determine the values of the parameters used for constructing a data-dependent feature set. We experimentally show that our approaches achieve the construction of informative feature sets for multiple datasets.

Keywords

Unsupervised feature construction Feature evaluation Nonparametric statistics Data mining Clustering Representations Algorithms for data and knowledge management Heuristic methods Pattern analysis 

Copyright information

© Springer Science+Business Media New York 2013

Authors and Affiliations

  • Marian-Andrei Rizoiu
    • 1
  • Julien Velcin
    • 1
  • Stéphane Lallich
    • 1
  1. 1.ERIC LaboratoryUniversity Lumière Lyon 2Bron CedexFrance

Personalised recommendations