Publication - Comparison of Regression Methods, Symbolic Induction Methods and

Authors: Sandholm. T., Brodley, C., Vidovic, A. and Sandholm, M.
Title: Comparison of Regression Methods, Symbolic Induction Methods and
Abstract: Classifier induction algorithms differ on what inductive hypotheses they can represent_ and on how they search their space of hypotheses. No classifier is better than another for all problems: they have selective superiority. This paper empirically compares six classifier induction algorithms on the diagnosis of equine colic and the prediction of its mortality. The classification is based on simultaneously analyzing sixteen features measured from a patient. The relative merits of the algorithms (linear regression, decision trees, nearest neighbor classifiers, the Model Class Selection system, logistic regression (with and without feature selection), and neural nets) are qualitatively discussed, and the generalization accuracies quantitatively analyzed.
Publication: Working Notes of the AAAI Spring, pp. 154 - 159
Date: January 1996
Sources: PS: ftp://ftp.cs.umass.edu/pub/lesser/sandholm-aaai96ss.ps
PDF: /Documents/sandholm-aaai96ss.pdf
Reference: Sandholm. T., Brodley, C., Vidovic, A. and Sandholm, M.. Comparison of Regression Methods, Symbolic Induction Methods and. Working Notes of the AAAI Spring, pp. 154-159. January 1996.
bibtex: 
@article{Sandholm-163,
  author    = "Brodley Sandholm. T. and Vidovic C. and Sandholm
               A. and M.",
  title     = "{Comparison of Regression Methods, Symbolic
               Induction Methods and}",
  journal   = "Working Notes of the AAAI Spring",
  pages     = "154-159",
  month     = "January",
  year      = "1996",
  url       = "http://mas.cs.umass.edu/paper/163",
}