Rough Set-based Dimensionality Reduction for Supervised and Unsupervised Learning

The curse of dimensionality is a damning factor for numerous potentially powerful machine learning techniques. Widely approved and otherwise elegant methodologies used for a number of different tasks ranging from classification to function approximation exhibit relatively high computational complexity with respect to dimensionality. This limits severely the applicability of such techniques to real world problems. Rough set theory is a formal methodology that can be employed to reduce the dimensionality of datasets as a preprocessing step to training a learning system on the data. This paper investigates the utility of the Rough Set Attribute Reduction (RSAR) technique to both supervised and unsupervised learning in an effort to probe RSAR’s generality. FuREAP, a Fuzzy-Rough Estimator of Algae Populations, which is an existing integration of RSAR and a fuzzy Rule Induction Algorithm (RIA), is used as an example of a supervised learning system with dimensionality reduction capabilities. A similar framework integrating the Multivariate Adaptive Regression Splines (MARS) approach and RSAR is taken to represent unsupervised learning systems. The paper describes the three techniques in question, discusses how RSAR can be employed with a supervised or an unsupervised system, and uses experimental results to draw conclusions on the relative success of the two integration efforts.