A classifier fitness measure based on Bayesian Likelilioods: an approach to the problem of learning from positives only

Authors:
Andrew Skabar;Anthony Maeder;Binh Pham
Affiliations:
School of Electrical & Electronic Systems Engineering, Queensland University of Technology, Brisbane, QLD, Australia;School of Electrical & Electronic Systems Engineering, Queensland University of Technology, Brisbane, QLD, Australia;Faculty of Information Technology, Queensland University of Technology, Brisbane, QLD, Australia
Venue:
PRICAI'00 Proceedings of the 6th Pacific Rim international conference on Artificial intelligence
Year:
2000

Citing 5
Cited 0

Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations

Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations
C4.5: programs for machine learning

C4.5: programs for machine learning
Using Genetic Algorithms for Concept Learning

Machine Learning - Special issue on genetic algorithms
Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control and Artificial Intelligence

Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control and Artificial Intelligence
Genetic Algorithms in Search, Optimization and Machine Learning

Genetic Algorithms in Search, Optimization and Machine Learning

Quantified Score

Hi-index	0.00

Visualization

Abstract

A classifier evaluation function based on Bayesian likelihoods of necessity and sufficiency is defined. This ftmction can be used to measure the performance of an arbitrary classifier on a set of examples consisting of labeled positives together with a corpus of unlabeled data. A neural network system has been implemented in which the evaluation function is used as a heuristic to guide search through the space of network weight configurations. Results are presented from testing the system on three artificial datasets. The results are comparable to those that can be obtained using back-propagation, despite the fact that the latter method requires labeled counter-examples.