Abstract

Empirical evidence shows that ensembles with adequate levels of pairwise diversity among a set of accurate member algorithms significantly outperform any of the individual algorithms. As a result, several diversity measures have been developed for use in optimizing ensembles. We show that diversity measures that properly combine the diversity space in an additive and multiplicative manner, not only result in ensembles whose accuracy is comparable to the naive ensemble of choosing the most accurate learners, but also results in ensembles that are significantly more efficient than such naive ensembles. In addition to diversity measures found in the literature, we submit two measures of diversity that span the diversity space in unique ways. Each of these measures considers not only the diversity of ratings between a pair of algorithms, but how this diversity relates to the target values.

Degree

College and Department

Physical and Mathematical Sciences; Computer Science

Rights

http://lib.byu.edu/about/copyright/