Comparative Study of Six Classification Methods for Mixtures of Variables,

Abstract

The performance of six discriminant methods is compared on simulated data consisting of mixtures of continuous, binary, ordinal and nominal variables. These methods are: Fisher's linear discrimination, logistic discrimination, quadratic discrimination, a kernel model, an independence model and the K-nearest neighbor method. In this paper, the simulation design was carefully conceived. The independence model with an association parameter performs well and is very robust.

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1992
Accession Number
ADP007140

Entities

People

  • O. Cherkaoui
  • R. Cleroux

Organizations

  • Université de Montréal

Tags

DTIC Thesaurus Topics

  • Classification
  • Computer Science
  • Data Science
  • Discrimination
  • Engineering
  • Information Science
  • Mathematics
  • Network Science
  • Simulations
  • Statistics
  • Theoretical Computer Science

Readers

  • Regression Analysis.
  • Statistical inference.