Domain based classification
This addresses a fundamental issue in classification for scenarios where traditional methods fail due to unreliable probability estimates, though it appears incremental as it builds on existing domain-based approaches.
The paper tackles the problem of classification when class probability distributions are ill-defined or impossible to estimate, proposing to use class domains instead of distributions to construct a decision function, with proposals for evaluation criteria and learning schemes illustrated by an example.
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.