Domain based classification

TL;DR

This paper proposes using class domains instead of probability distributions for classification when class distributions are ill-defined, offering new evaluation criteria and learning schemes to improve decision reliability.

Contribution

It introduces a novel approach of using class domains for classification, addressing issues with traditional distribution-based methods when distributions are ill-defined.

Findings

Class domains provide a reliable alternative to class distributions.

New evaluation criteria for domain-based classifiers are proposed.

An example demonstrates the effectiveness of the approach.

Abstract

The majority of traditional classification ru les minimizing the expected probability of error (0-1 loss) are inappropriate if the class probability distributions are ill-defined or impossible to estimate. We argue that in such cases class domains should be used instead of class distributions or densities to construct a reliable decision function. Proposals are presented for some evaluation criteria and classifier learning schemes, illustrated by an example.