Class-specific feature selection based on uniform dirichlet priors
Robert S. Lynch, Peter Willett · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2000
In this paper, the Bayesian Data Reduction Algorithm (BDRA) is applied to reducing the dimensionality of a data set that contains class-specific feature. The BDRA uses the probability of error, conditioned on the training data, and a 'greedy' approach for reducing irrelevant features from the data. Here, the BDRA is shown to be an effective means of selecting binary valued class-specific feature, where the remaining non-class-specific features are irrelevant to correct classification. In fact, performance results reveal that when using a small number of training data relative to feature dimensionality, the BDRA outperforms the appropriate class-specific classifier.